File size: 225,203 Bytes
6fa4bc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
1393
1394
1395
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411
1412
1413
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611
1612
1613
1614
1615
1616
1617
1618
1619
1620
1621
1622
1623
1624
1625
1626
1627
1628
1629
1630
1631
1632
1633
1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
1649
1650
1651
1652
1653
1654
1655
1656
1657
1658
1659
1660
1661
1662
1663
1664
1665
1666
1667
1668
1669
1670
1671
1672
1673
1674
1675
1676
1677
1678
1679
1680
1681
1682
1683
1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
1703
1704
1705
1706
1707
1708
1709
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743
1744
1745
1746
1747
1748
1749
1750
1751
1752
1753
1754
1755
1756
1757
1758
1759
1760
1761
1762
1763
1764
1765
1766
1767
1768
1769
1770
1771
1772
1773
1774
1775
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899
1900
1901
1902
1903
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914
1915
1916
1917
1918
1919
1920
1921
1922
1923
1924
1925
1926
1927
1928
1929
1930
1931
1932
1933
1934
1935
1936
1937
1938
1939
1940
1941
1942
1943
1944
1945
1946
1947
1948
1949
1950
1951
1952
1953
1954
1955
1956
1957
1958
1959
1960
1961
1962
1963
1964
1965
1966
1967
1968
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
2015
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049
2050
2051
2052
2053
2054
2055
2056
2057
2058
2059
2060
2061
2062
2063
2064
2065
2066
2067
2068
2069
2070
2071
2072
2073
2074
2075
2076
2077
2078
2079
2080
2081
2082
2083
2084
2085
2086
2087
2088
2089
2090
2091
2092
2093
2094
2095
2096
2097
2098
2099
2100
2101
2102
2103
2104
2105
2106
2107
2108
2109
2110
2111
2112
2113
2114
2115
2116
2117
2118
2119
2120
2121
2122
2123
2124
2125
2126
2127
2128
2129
2130
2131
2132
2133
2134
2135
2136
2137
2138
2139
2140
2141
2142
2143
2144
2145
2146
2147
2148
2149
2150
2151
2152
2153
2154
2155
2156
2157
2158
2159
2160
2161
2162
2163
2164
2165
2166
2167
2168
2169
2170
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187
2188
2189
2190
2191
2192
2193
2194
2195
2196
2197
2198
2199
2200
2201
2202
2203
2204
2205
2206
2207
2208
2209
2210
2211
2212
2213
2214
2215
2216
2217
2218
2219
2220
2221
2222
2223
2224
2225
2226
2227
2228
2229
2230
2231
2232
2233
2234
2235
2236
2237
2238
2239
2240
2241
2242
2243
2244
2245
2246
2247
2248
2249
2250
2251
2252
2253
2254
2255
2256
2257
2258
2259
2260
2261
2262
2263
2264
2265
2266
2267
2268
2269
2270
2271
2272
2273
2274
2275
2276
2277
2278
2279
2280
2281
2282
2283
2284
2285
2286
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306
2307
2308
2309
2310
2311
2312
2313
2314
2315
2316
2317
2318
2319
2320
2321
2322
2323
2324
2325
2326
2327
2328
2329
2330
2331
2332
2333
2334
2335
2336
2337
2338
2339
2340
2341
2342
2343
2344
2345
2346
2347
2348
2349
2350
2351
2352
2353
2354
2355
2356
2357
2358
2359
2360
2361
2362
2363
2364
2365
2366
2367
2368
2369
2370
2371
2372
2373
2374
2375
2376
2377
2378
2379
2380
2381
2382
2383
2384
2385
2386
2387
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398
2399
2400
2401
2402
2403
2404
2405
2406
2407
2408
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418
2419
2420
2421
2422
2423
2424
2425
2426
2427
2428
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438
2439
2440
2441
2442
2443
2444
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457
2458
2459
2460
2461
2462
2463
2464
2465
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478
2479
2480
2481
2482
2483
2484
2485
2486
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498
2499
2500
2501
2502
2503
2504
2505
2506
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518
2519
2520
2521
2522
2523
2524
2525
2526
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549
2550
2551
2552
2553
2554
2555
2556
2557
2558
2559
2560
2561
2562
2563
2564
2565
2566
2567
2568
2569
2570
2571
2572
2573
2574
2575
2576
2577
2578
2579
2580
2581
2582
2583
2584
2585
2586
2587
2588
2589
2590
2591
2592
2593
2594
2595
2596
2597
2598
2599
2600
2601
2602
2603
2604
2605
2606
2607
2608
2609
2610
2611
2612
2613
2614
2615
2616
2617
2618
2619
2620
2621
2622
2623
2624
2625
2626
2627
2628
2629
2630
2631
2632
2633
2634
2635
2636
2637
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657
2658
2659
2660
2661
2662
2663
2664
2665
2666
2667
2668
2669
2670
2671
2672
2673
2674
2675
2676
2677
2678
2679
2680
2681
2682
2683
2684
2685
2686
2687
2688
2689
2690
2691
2692
2693
2694
2695
2696
2697
2698
2699
2700
2701
2702
2703
2704
2705
2706
2707
2708
2709
2710
2711
2712
2713
2714
2715
2716
2717
2718
2719
2720
2721
2722
2723
2724
2725
2726
2727
2728
2729
2730
2731
2732
2733
2734
2735
2736
2737
2738
2739
2740
2741
2742
2743
2744
2745
2746
2747
2748
2749
2750
2751
2752
2753
2754
2755
2756
2757
2758
2759
2760
2761
2762
2763
2764
2765
2766
2767
2768
2769
2770
2771
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783
2784
2785
2786
2787
2788
2789
2790
2791
2792
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819
2820
2821
2822
2823
2824
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835
2836
2837
2838
2839
2840
2841
2842
2843
2844
2845
2846
2847
2848
2849
2850
2851
2852
2853
2854
2855
2856
2857
2858
2859
2860
2861
2862
2863
2864
2865
2866
2867
2868
2869
2870
2871
2872
2873
2874
2875
2876
2877
2878
2879
2880
2881
2882
2883
2884
2885
2886
2887
2888
2889
2890
2891
2892
2893
2894
2895
2896
2897
2898
2899
2900
2901
2902
2903
2904
2905
2906
2907
2908
2909
2910
2911
2912
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936
2937
2938
2939
2940
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953
2954
2955
2956
2957
2958
2959
2960
2961
2962
2963
2964
2965
2966
2967
2968
2969
2970
2971
2972
2973
2974
2975
2976
2977
2978
2979
2980
2981
2982
2983
2984
2985
2986
2987
2988
2989
2990
2991
2992
2993
2994
2995
2996
2997
2998
2999
3000
3001
3002
3003
3004
3005
3006
3007
3008
3009
3010
3011
3012
3013
3014
3015
3016
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033
3034
3035
3036
3037
3038
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053
3054
3055
3056
3057
3058
3059
3060
3061
3062
3063
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073
3074
3075
3076
3077
3078
3079
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091
3092
3093
3094
3095
3096
3097
3098
3099
3100
3101
3102
3103
3104
3105
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118
3119
3120
3121
3122
3123
3124
3125
3126
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139
3140
3141
3142
3143
3144
3145
3146
3147
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159
3160
3161
3162
3163
3164
3165
3166
3167
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179
3180
3181
3182
3183
3184
3185
3186
3187
3188
3189
3190
3191
3192
3193
3194
3195
3196
3197
3198
3199
3200
3201
3202
3203
3204
3205
3206
3207
3208
3209
3210
3211
3212
3213
3214
3215
3216
3217
3218
3219
3220
3221
3222
3223
3224
3225
3226
3227
3228
3229
3230
3231
3232
3233
3234
3235
3236
3237
3238
3239
3240
3241
3242
3243
3244
3245
3246
3247
3248
3249
3250
3251
3252
3253
3254
3255
3256
3257
3258
3259
3260
3261
3262
3263
3264
3265
3266
3267
3268
3269
3270
3271
3272
3273
3274
3275
3276
3277
3278
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298
3299
3300
3301
3302
3303
3304
3305
3306
3307
3308
3309
3310
3311
3312
3313
3314
3315
3316
3317
3318
3319
3320
3321
3322
3323
3324
3325
3326
3327
3328
3329
3330
3331
3332
3333
3334
3335
3336
3337
3338
3339
3340
3341
3342
3343
3344
3345
3346
3347
3348
3349
3350
3351
3352
3353
3354
3355
3356
3357
3358
3359
3360
3361
3362
3363
3364
3365
3366
3367
3368
3369
3370
3371
3372
3373
3374
3375
3376
3377
3378
3379
3380
3381
3382
3383
3384
3385
3386
3387
3388
3389
3390
3391
3392
3393
3394
3395
3396
3397
3398
3399
3400
3401
3402
3403
3404
3405
3406
3407
3408
3409
3410
3411
3412
3413
3414
3415
3416
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428
3429
3430
3431
3432
3433
3434
3435
3436
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464
3465
3466
3467
3468
3469
3470
3471
3472
3473
3474
3475
3476
3477
3478
3479
3480
3481
3482
3483
3484
3485
3486
3487
3488
3489
3490
3491
3492
3493
3494
3495
3496
3497
3498
3499
3500
3501
3502
3503
3504
3505
3506
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
{
    "paper_id": "J09-3003",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T02:53:59.210589Z"
    },
    "title": "Recognizing Contextual Polarity: An Exploration of Features for Phrase-Level Sentiment Analysis",
    "authors": [
        {
            "first": "Theresa",
            "middle": [],
            "last": "Wilson",
            "suffix": "",
            "affiliation": {},
            "email": "twilson@inf.ed.ac.uk."
        },
        {
            "first": "Janyce",
            "middle": [],
            "last": "Wiebe",
            "suffix": "",
            "affiliation": {},
            "email": "wiebe@cs.pitt.edu"
        },
        {
            "first": "Paul",
            "middle": [],
            "last": "Hoffmann",
            "suffix": "",
            "affiliation": {},
            "email": "hoffmanp@cs.pitt.edu"
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Many approaches to automatic sentiment analysis begin with a large lexicon of words marked with their prior polarity (also called semantic orientation). However, the contextual polarity of the phrase in which a particular instance of a word appears may be quite different from the word's prior polarity. Positive words are used in phrases expressing negative sentiments, or vice versa. Also, quite often words that are positive or negative out of context are neutral in context, meaning they are not even being used to express a sentiment. The goal of this work is to automatically distinguish between prior and contextual polarity, with a focus on understanding which features are important for this task. Because an important aspect of the problem is identifying when polar terms are being used in neutral contexts, features for distinguishing between neutral and polar instances are evaluated, as well as features for distinguishing between positive and negative contextual polarity. The evaluation includes assessing the performance of features across multiple machine learning algorithms. For all learning algorithms except one, the combination of all features together gives the best performance. Another facet of the evaluation considers how the presence of neutral instances affects the performance of features for distinguishing between positive and negative polarity. These experiments show that the presence of neutral instances greatly degrades the performance of these features, and that perhaps the best way to improve performance across all polarity classes is to improve the system's ability to identify when an instance is neutral.",
    "pdf_parse": {
        "paper_id": "J09-3003",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Many approaches to automatic sentiment analysis begin with a large lexicon of words marked with their prior polarity (also called semantic orientation). However, the contextual polarity of the phrase in which a particular instance of a word appears may be quite different from the word's prior polarity. Positive words are used in phrases expressing negative sentiments, or vice versa. Also, quite often words that are positive or negative out of context are neutral in context, meaning they are not even being used to express a sentiment. The goal of this work is to automatically distinguish between prior and contextual polarity, with a focus on understanding which features are important for this task. Because an important aspect of the problem is identifying when polar terms are being used in neutral contexts, features for distinguishing between neutral and polar instances are evaluated, as well as features for distinguishing between positive and negative contextual polarity. The evaluation includes assessing the performance of features across multiple machine learning algorithms. For all learning algorithms except one, the combination of all features together gives the best performance. Another facet of the evaluation considers how the presence of neutral instances affects the performance of features for distinguishing between positive and negative polarity. These experiments show that the presence of neutral instances greatly degrades the performance of these features, and that perhaps the best way to improve performance across all polarity classes is to improve the system's ability to identify when an instance is neutral.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "Sentiment analysis is a type of subjectivity analysis (Wiebe 1994 ) that focuses on identifying positive and negative opinions, emotions, and evaluations expressed in natural language. It has been a central component in applications ranging from recognizing inflammatory messages (Spertus 1997) , to tracking sentiments over time in online discussions (Tong 2001) , to classifying positive and negative reviews (Pang, Lee, and Vaithyanathan 2002; Turney 2002) . Although a great deal of work in sentiment analysis has targeted documents, applications such as opinion question answering (Yu and Hatzivassiloglou 2003; Maybury 2004; Stoyanov, Cardie, and Wiebe 2005) and review mining to extract opinions about companies and products (Morinaga et al. 2002; ) require sentence-level or even phrase-level analysis. For example, if a question answering system is to successfully answer questions about people's opinions, it must be able not only to pinpoint expressions of positive and negative sentiments, such as we find in sentence (1), but also to determine when an opinion is not being expressed by a word or phrase that typically does evoke one, such as condemned in sentence (2).",
                "cite_spans": [
                    {
                        "start": 54,
                        "end": 65,
                        "text": "(Wiebe 1994",
                        "ref_id": "BIBREF37"
                    },
                    {
                        "start": 280,
                        "end": 294,
                        "text": "(Spertus 1997)",
                        "ref_id": "BIBREF28"
                    },
                    {
                        "start": 352,
                        "end": 363,
                        "text": "(Tong 2001)",
                        "ref_id": "BIBREF33"
                    },
                    {
                        "start": 411,
                        "end": 446,
                        "text": "(Pang, Lee, and Vaithyanathan 2002;",
                        "ref_id": null
                    },
                    {
                        "start": 447,
                        "end": 459,
                        "text": "Turney 2002)",
                        "ref_id": "BIBREF34"
                    },
                    {
                        "start": 586,
                        "end": 616,
                        "text": "(Yu and Hatzivassiloglou 2003;",
                        "ref_id": "BIBREF45"
                    },
                    {
                        "start": 617,
                        "end": 630,
                        "text": "Maybury 2004;",
                        "ref_id": "BIBREF22"
                    },
                    {
                        "start": 631,
                        "end": 664,
                        "text": "Stoyanov, Cardie, and Wiebe 2005)",
                        "ref_id": "BIBREF30"
                    },
                    {
                        "start": 732,
                        "end": 754,
                        "text": "(Morinaga et al. 2002;",
                        "ref_id": "BIBREF23"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "(1) African observers generally approved (positive) of his victory while Western governments denounced (negative) it.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "(2) Gavin Elementary School was condemned in April 2004.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "A common approach to sentiment analysis is to use a lexicon with information about which words and phrases are positive and which are negative. This lexicon may be manually compiled, as is the case with the General Inquirer (Stone et al. 1966) , a resource often used in sentiment analysis. Alternatively, the information in the lexicon may be acquired automatically. Acquiring the polarity of words and phrases is itself an active line of research in the sentiment analysis community, pioneered by the work of Hatzivassiloglou and McKeown (1997) on predicting the polarity or semantic orientation of adjectives. Various techniques have been proposed for learning the polarity of words. They include corpus-based techniques, such as using constraints on the co-occurrence in conjunctions of words with similar or opposite polarity (Hatzivassiloglou and McKeown 1997) and statistical measures of word association (Turney and Littman 2003) , as well as techniques that exploit information about lexical relationships (Kamps and Marx 2002; Kim and Hovy 2004) and glosses (Esuli and Sebastiani 2005; Andreevskaia and Bergler 2006) in resources such as WordNet.",
                "cite_spans": [
                    {
                        "start": 224,
                        "end": 243,
                        "text": "(Stone et al. 1966)",
                        "ref_id": "BIBREF29"
                    },
                    {
                        "start": 511,
                        "end": 546,
                        "text": "Hatzivassiloglou and McKeown (1997)",
                        "ref_id": null
                    },
                    {
                        "start": 831,
                        "end": 866,
                        "text": "(Hatzivassiloglou and McKeown 1997)",
                        "ref_id": null
                    },
                    {
                        "start": 912,
                        "end": 937,
                        "text": "(Turney and Littman 2003)",
                        "ref_id": "BIBREF35"
                    },
                    {
                        "start": 1037,
                        "end": 1055,
                        "text": "Kim and Hovy 2004)",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 1068,
                        "end": 1095,
                        "text": "(Esuli and Sebastiani 2005;",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 1096,
                        "end": 1126,
                        "text": "Andreevskaia and Bergler 2006)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "Acquiring the polarity of words and phrases is undeniably important, and there are still open research challenges, such as addressing the sentiments of different senses of words (Esuli and Sebastiani 2006b; Wiebe and Mihalcea 2006) , and so on. However, what the polarity of a given word or phrase is when it is used in a particular context is another problem entirely. Consider, for example, the underlined positive and negative words in the following sentence.",
                "cite_spans": [
                    {
                        "start": 178,
                        "end": 206,
                        "text": "(Esuli and Sebastiani 2006b;",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 207,
                        "end": 231,
                        "text": "Wiebe and Mihalcea 2006)",
                        "ref_id": "BIBREF39"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "(3) Philip Clapp, president of the National Environment Trust, sums up well the general thrust of the reaction of environmental movements: \"There is no reason at all to believe that the polluters are suddenly going to become reasonable.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "The first underlined word is Trust. Although many senses of the word trust express a positive sentiment, in this case, the word is not being used to express a sentiment at all. It is simply part of an expression referring to an organization that has taken on the charge of caring for the environment. The adjective well is considered positive, and indeed it is positive in this context. However, the same is not true for the words reason and reasonable. Out of context, we would consider both of these words to be positive. 1 In context, the word reason is being negated, changing its polarity from positive to negative. The phrase no reason at all to believe changes the polarity of the proposition that follows; because reasonable falls within this proposition, its polarity becomes negative.",
                "cite_spans": [
                    {
                        "start": 524,
                        "end": 525,
                        "text": "1",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "The word polluters has a negative connotation, but here in the context of the discussion of the article and its position in the sentence, polluters is being used less to express a sentiment and more to objectively refer to companies that pollute. To clarify how the polarity of polluters is affected by its subject role, consider the purely negative sentiment that emerges when it is used as an object: They are polluters.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "We call the polarity that would be listed for a word in a lexicon the word's prior polarity, and we call the polarity of the expression in which a word appears, considering the context of the sentence and document, the word's contextual polarity. Although words often do have the same prior and contextual polarity, many times a word's prior and contextual polarities differ. Words with a positive prior polarity may have a negative contextual polarity, or vice versa. Quite often words that are positive or negative out of context are neutral in context, meaning that they are not even being used to express a sentiment. Similarly, words that are neutral out of context, neither positive or negative, may combine to create a positive or negative expression in context.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "The focus of this work is on the recognition of contextual polarity-in particular, disambiguating the contextual polarity of words with positive or negative prior polarity. We begin by presenting an annotation scheme for marking sentiment expressions and their contextual polarity in the Multi-perspective Question Answering (MPQA) opinion corpus. We show that, given a set of subjective expressions (identified from the existing annotations in the MPQA corpus), contextual polarity can be annotated reliably.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "Using the contextual polarity annotations, we conduct experiments in automatically distinguishing between prior and contextual polarity. Beginning with a large lexicon of clues tagged with prior polarity, we identify the contextual polarity of the instances of those clues in the corpus. The process that we use has two steps, first classifying each clue as being in a neutral or polar phrase, and then disambiguating the contextual polarity of the clues marked as polar. For each step in the process, we experiment with a variety of features and evaluate the performance of the features using several different machine learning algorithms.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "Our experiments reveal a number of interesting findings. First, being able to accurately identify neutral contextual polarity-when a positive or negative clue is not being used to express a sentiment-is an important aspect of the problem. The importance of neutral examples has previously been noted for classifying the sentiment of documents (Koppel and Schler 2006) , but ours is the first work to explore how neutral instances affect classifying the contextual polarity of words and phrases. In particular, we found that the performance of features for distinguishing between positive and negative polarity greatly degrades when neutral instances are included in the experiments.",
                "cite_spans": [
                    {
                        "start": 343,
                        "end": 367,
                        "text": "(Koppel and Schler 2006)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "We also found that achieving the best performance for recognizing contextual polarity requires a wide variety of features. This is particularly true for distinguishing between neutral and polar instances. Although some features help to increase polar or neutral recall or precision, it is only the combination of features together that achieves significant improvements in accuracy over the baselines. Our experiments show that for distinguishing between positive and negative instances, features capturing negation are clearly the most important. However, there is more to the story than simple negation. Features that capture relationships between instances of clues also perform well, indicating that identifying features that represent more complex interdependencies between sentiment clues may be an important avenue for future research.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "The remainder of this article is organized as follows. Section 2 gives an overview of some of the things that can influence contextual polarity. In Section 3, we describe our corpus and present our annotation scheme and inter-annotator agreement study for marking contextual polarity. Sections 4 and 5 describe the lexicon used in our experiments and how the contextual polarity annotations are used to determine the gold-standard tags for instances from the lexicon. In Section 6, we consider what kind of performance can be expected from a simple, prior-polarity classifier. Section 7 describes the features that we use for recognizing contextual polarity, and our experiments and results are presented in Section 8. In Section 9 we discuss related work, and we conclude in Section 10.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1."
            },
            {
                "text": "Phrase-level sentiment analysis is not a simple problem. Many things besides negation can influence contextual polarity, and even negation is not always straightforward. Negation may be local (e.g., not good), or involve longer-distance dependencies such as the negation of the proposition (e.g., does not look very good) or the negation of the subject (e.g., no one thinks that it's good). In addition, certain phrases that contain negation words intensify rather than change polarity (e.g., not only good but amazing). Contextual polarity may also be influenced by modality: whether the proposition is asserted to be real (realis) or not real (irrealis) (no reason at all to believe is irrealis, for example); word sense (e.g., Environmental Trust vs. He has won the people's trust); the syntactic role of a word in the sentence: whether the word is the subject or object of a copular verb (consider polluters are versus they are polluters); and diminishers such as little (e.g., little truth, little threat). Polanyi and Zaenen (2004) give a detailed discussion of many of these types of polarity influencers. Many of these contextual polarity influencers are represented as features in our experiments.",
                "cite_spans": [
                    {
                        "start": 1012,
                        "end": 1037,
                        "text": "Polanyi and Zaenen (2004)",
                        "ref_id": "BIBREF25"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Polarity Influencers",
                "sec_num": "2."
            },
            {
                "text": "Contextual polarity may also be influenced by the domain or topic. For example, the word cool is positive if used to describe a car, but it is negative if it is used to describe someone's demeanor. Similarly, a word such as fever is unlikely to be expressing a sentiment when used in a medical context. We use one feature in our experiments to represent the topic of the document.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Polarity Influencers",
                "sec_num": "2."
            },
            {
                "text": "Another important aspect of contextual polarity is the perspective of the person who is expressing the sentiment. For example, consider the phrase failed to defeat in the sentence Israel failed to defeat Hezbollah. From the perspective of Israel, failed to defeat is negative. From the perspective of Hezbollah, failed to defeat is positive. Therefore, the contextual polarity of this phrase ultimately depends on the perspective of who is expressing the sentiment. Although automatically detecting this kind of pragmatic influence on polarity is beyond the scope of this work, this as well as the other types of polarity influencers all are considered when annotating contextual polarity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Polarity Influencers",
                "sec_num": "2."
            },
            {
                "text": "For the experiments in this work, we need a corpus that is annotated comprehensively for sentiment expressions and their contextual polarity. Rather than building a corpus from scratch, we chose to add contextual polarity annotations to the existing annotations in the Multi-perspective Question Answering (MPQA) opinion corpus 2 (Wiebe, Wilson, and Cardie 2005) .",
                "cite_spans": [
                    {
                        "start": 330,
                        "end": 362,
                        "text": "(Wiebe, Wilson, and Cardie 2005)",
                        "ref_id": "BIBREF41"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data and Annotations",
                "sec_num": "3."
            },
            {
                "text": "The MPQA corpus is a collection of English-language versions of news documents from the world press. The documents contain detailed, expression-level annotations of attributions and private states (Quirk et al. 1985) . Private states are mental and emotional states; they include beliefs, speculations, intentions, and sentiments, among others. Although sentiments are not distinguished from other types of private states in the existing annotations, they are a subset of what already is annotated. This makes the annotations in the MPQA corpus a good starting point for annotating sentiment expressions and their contextual polarity.",
                "cite_spans": [
                    {
                        "start": 197,
                        "end": 216,
                        "text": "(Quirk et al. 1985)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data and Annotations",
                "sec_num": "3."
            },
            {
                "text": "When developing our annotation scheme for sentiment expressions and contextual polarity, there were three main questions to address. First, which of the existing annotations in the MPQA corpus have the possibility of being sentiment expressions? Second, which of the possible sentiment expressions actually are expressing sentiments? Third, what coding scheme should be used for marking contextual polarity?",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "The MPQA annotation scheme has four types of annotations: objective speech event frames, two types of private state frames, and agent frames that are used for marking speakers of speech events and experiencers of private states. A full description of the MPQA annotation scheme and an agreement study evaluating key aspects of the scheme are found in Wiebe, Wilson, and Cardie (2005) .",
                "cite_spans": [
                    {
                        "start": 351,
                        "end": 383,
                        "text": "Wiebe, Wilson, and Cardie (2005)",
                        "ref_id": "BIBREF41"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "The two types of private state frames, direct subjective frames and expressive subjective element frames, are where we will find sentiment expressions. Direct subjective frames are used to mark direct references to private states as well as speech events in which private states are being expressed. For example, in the following sentences, fears, praised, and said are all marked as direct subjective annotations.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "(4) The U.S. fears a spill-over of the anti-terrorist campaign.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "(5) Italian senator Renzo Gubert praised the Chinese government's efforts.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "(6) \"The report is full of absurdities,\" he said.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "The word fears directly refers to a private state; praised refers to a speech event in which a private state is being expressed; and said is marked as direct subjective because a private state is being expressed within the speech event referred to by said. Expressive subjective elements indirectly express private states through the way something is described or through a particular wording. In example (6), the phrase full of absurdities is an expressive subjective element. Subjectivity (Banfield 1982; Wiebe 1994 ) refers to the linguistic expression of private states, hence the names for the two types of private state annotations.",
                "cite_spans": [
                    {
                        "start": 491,
                        "end": 506,
                        "text": "(Banfield 1982;",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 507,
                        "end": 517,
                        "text": "Wiebe 1994",
                        "ref_id": "BIBREF37"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "All expressive subjective elements are included in the set of annotations that have the possibility of being sentiment expressions, but the direct subjective frames to include in this set can be pared down further. Direct subjective frames have an attribute, expression intensity, that captures the contribution of the annotated word or phrase to the overall intensity of the private state being expressed. Expression intensity ranges from neutral to high. In the given sentences, fears and praised have an expression intensity of medium, and said has an expression intensity of neutral. A neutral expression intensity indicates that the direct subjective phrase itself is not contributing to the expression of the private state. If this is the case, then the direct subjective phrase cannot be a sentiment expression. Thus, only direct subjective annotations with a non-neutral expression intensity are included in the set of annotations that have the possibility of being sentiment expressions. We call this set of annotations, the union of the expressive subjective elements and the direct subjective frames with a non-neutral intensity, the subjective expressions in the corpus; these are the annotations we will mark for contextual polarity. Table 1 gives a sample of subjective expressions marked in the MPQA corpus. Although many of the words and phrases express what we typically think of as sentiments, others do not, for example, believes, very definitely, and unconditionally and without delay.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1247,
                        "end": 1254,
                        "text": "Table 1",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "Now that we have identified which annotations have the possibility of being sentiment expressions, the next question is which of these annotated words and phrases are actually expressing sentiments. We define a sentiment as a positive or negative emotion, evaluation, or stance. On the left of Table 2 are examples of positive sentiments; examples of negative sentiments are on the right. The final issue to address is the actual annotation scheme for marking contextual polarity. The scheme we developed has four tags: positive, negative, both, and neutral. The positive tag is used to mark positive sentiments. The negative tag is used to mark negative sentiments. The both tag is applied to expressions in which both a positive and negative sentiment are being expressed. Subjective expressions with positive, negative, or both tags are our sentiment expressions. The neutral tag is used for all other subjective expressions, including emotions, evaluations, and stances that are neither positive or negative. Instructions for the contextual-polarity annotation scheme are available at http://www.cs.pitt.edu/mpqa/databaserelease/polarityCodingInstructions.txt.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 294,
                        "end": 301,
                        "text": "Table 2",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "Following are examples from the corpus of each of the different contextual-polarity annotations. Each underlined word or phrase is a subjective expression that was marked in the original MPQA annotations. 3 In bold following each subjective expression is the contextual polarity with which it was annotated. (10) Jerome says the hospital feels (neutral) no different than a hospital in the states.",
                "cite_spans": [
                    {
                        "start": 205,
                        "end": 206,
                        "text": "3",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "As a final note on the annotation scheme, annotators are asked to judge the contextual polarity of the sentiment that is ultimately being conveyed by the subjective expression, that is, once the sentence has been fully interpreted. Thus, the subjective expression, they have not succeeded, and will never succeed, is marked as positive in the following sentence:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "(11) They have not succeeded, and will never succeed (positive), in breaking the will of this valiant people.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "The reasoning is that breaking the will of a valiant people is negative, so to not succeed in breaking their will is positive. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Annotation Scheme",
                "sec_num": "3.1"
            },
            {
                "text": "To measure the reliability of the polarity annotation scheme, we conducted an agreement study with two annotators 4 using 10 documents from the MPQA corpus. The 10 documents contain 447 subjective expressions. Table 3 shows the contingency table for the two annotators' judgments. Overall agreement is 82%, with a kappa value of 0.72. As part of the annotation scheme, annotators are asked to judge how certain they are in their polarity tags. For 18% of the subjective expressions, at least one annotator used the uncertain tag when marking polarity. If we consider these cases to be borderline and exclude them from the study, percent agreement increases to 90% and kappa rises to 0.84. Table 4 shows the revised contingency table with the uncertain cases removed. This shows that annotator agreement is especially high when both annotators are certain, and that annotators are certain for over 80% of their tags.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 210,
                        "end": 217,
                        "text": "Table 3",
                        "ref_id": "TABREF2"
                    },
                    {
                        "start": 689,
                        "end": 696,
                        "text": "Table 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Agreement Study",
                "sec_num": "3.2"
            },
            {
                "text": "Note that all annotations are included in the experiments.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Agreement Study",
                "sec_num": "3.2"
            },
            {
                "text": "In total, all 19,962 subjective expressions in the 535 documents (11,112 sentences) of the MPQA corpus were annotated with their contextual polarity as just described. 5 Three annotators carried out the task: the two who participated in the annotation study and a third who was trained later. 6 Table 5 gives the distribution of the contextual polarity tags. Looking at this table, we see that a small majority of subjective expressions (54.6%) are expressing a positive, negative, or both (positive and negative) sentiment. We refer to these expressions as polar in context. Many of the subjective expressions are neutral and do not express a sentiment. This suggests that, although sentiment is a major type of subjectivity, distinguishing other prominent types of subjectivity will be important for future work in subjectivity analysis. As many NLP applications operate at the sentence level, one important issue to consider is the distribution of sentences with respect to the subjective expressions they contain. In the 11,112 sentences in the MPQA corpus, 28% contain no subjective expressions, 24% contain only one, and 48% contain two or more. Of the 5,304 sentences containing two or more subjective expressions, 17% contain mixtures of positive and negative expressions, and 61% contain mixtures of polar (positive/negative/both) and neutral subjective expressions.",
                "cite_spans": [
                    {
                        "start": 168,
                        "end": 169,
                        "text": "5",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 295,
                        "end": 302,
                        "text": "Table 5",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Contextual Polarity Annotations",
                "sec_num": "3.3"
            },
            {
                "text": "For the experiments in this article, we use a lexicon of over 8,000 subjectivity clues. Subjectivity clues are words and phrases that may be used to express private states. In other words, subjectivity clues have subjective usages, though they may have objective usages as well. For this work, only single-word clues are used.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "To compile the lexicon, we began with the list of subjectivity clues from Riloff and Wiebe (2003) , which includes the positive and negative adjectives from Hatzivassiloglou and McKeown (1997) . The words in this list were grouped in previous work according to their reliability as subjectivity clues. Words that are subjective in most contexts are considered strong subjective clues, indicated by the strongsubj tag. Words that may only have certain subjective usages are considered weak subjective clues, indicated by the weaksubj tag.",
                "cite_spans": [
                    {
                        "start": 74,
                        "end": 97,
                        "text": "Riloff and Wiebe (2003)",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 157,
                        "end": 192,
                        "text": "Hatzivassiloglou and McKeown (1997)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "We expanded the list using a dictionary and a thesaurus, and added words from the General Inquirer positive and negative word lists (Stone et al. 1966 ) that we judged to be potentially subjective. 7 We also gave the new words strongsubj and weaksubj reliability tags. The final lexicon has a coverage of 67% of subjective expressions in the MPQA corpus, where coverage is the percentage of subjective expressions containing one or more instances of clues from the lexicon. The coverage of just sentiment expressions is even higher: 75%.",
                "cite_spans": [
                    {
                        "start": 132,
                        "end": 150,
                        "text": "(Stone et al. 1966",
                        "ref_id": "BIBREF29"
                    },
                    {
                        "start": 198,
                        "end": 199,
                        "text": "7",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "The next step was to tag the clues in the lexicon with their prior polarity: positive, negative, both, or neutral. A word in the lexicon is tagged as positive if out of context it seems to evoke something positive, and negative if it seems to evoke something negative. If a word has both positive and negative meanings, it is tagged with the polarity that seems the most common. A word is tagged as both if it is at the same time both positive and negative. For example, the word bittersweet evokes something both positive and negative. Words like brag are also tagged as both, because the one who is bragging is expressing something positive, yet at the same time describing someone as bragging is expressing a negative evaluation of that person. A word is tagged as neutral if it does not evoke anything positive or negative.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "For words that came from positive and negative word lists (Stone et al. 1966; Hatzivassiloglou and McKeown 1997) , we largely retained their original polarity. However, we did change the polarity of a word if we strongly disagreed with its original class. 8 For example, the word apocalypse is listed as positive in the General Inquirer; we changed its prior polarity to negative for our lexicon.",
                "cite_spans": [
                    {
                        "start": 58,
                        "end": 77,
                        "text": "(Stone et al. 1966;",
                        "ref_id": "BIBREF29"
                    },
                    {
                        "start": 78,
                        "end": 112,
                        "text": "Hatzivassiloglou and McKeown 1997)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "By far, the majority of clues in the lexicon (92.8%) are marked as having either positive (33.1%) or negative (59.7%) prior polarity. Only a small number of clues (0.3%) are marked as having both positive and negative polarity. We refer to the set of clues marked as positive, negative, or both as sentiment clues. A total of 6.9% of the clues in the lexicon are marked as neutral. Examples of neutral clues are verbs such as feel, look, and think, and intensifiers such as deeply, entirely, and practically. Although the neutral clues make up a small proportion of the total words in the lexicon, we retain them for our later experiments in recognizing contextual polarity because many of them are good clues that a sentiment is being expressed (e.g., feels slighted, feels satisfied, look kindly on, look forward to). Including them increases the coverage of the system.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "At the end of the previous section, we considered the distribution of sentences in the MPQA corpus with respect to the subjective expressions they contain. It is interesting to compare that distribution with the distribution of sentences with respect to the instances they contain of clues from the lexicon. We find that there are more sentences with two or more clue instances (62%) than sentences with two or more subjective expressions (48%). More importantly, many more sentences have mixtures of positive and negative clue instances than actually have mixtures of positive and negative subjective expressions. Only 880 sentences have a mixture of both positive and negative subjective expressions, whereas 3,234 sentences have a mixture of positive and negative clue instances. Thus, a large number of positive and negative instances are either neutral in context, or they are combining to form more complex polarity expressions. Either way, this provides strong evidence of the need to be able to disambiguate the contextual polarity of subjectivity and sentiment clues.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Prior-Polarity Subjectivity Lexicon",
                "sec_num": "4."
            },
            {
                "text": "In the experiments described in the following sections, the goal is to classify the contextual polarity of the expressions that contain instances of the subjectivity clues in our lexicon. However, determining which clue instances are part of the same expression and identifying expression boundaries are not the focus of this work. Thus, instead of trying to identify and label each expression, in the following experiments, each clue instance is labeled individually as to its contextual polarity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Definition of the Gold Standard",
                "sec_num": "5."
            },
            {
                "text": "We define the gold-standard contextual polarity of a clue instance in terms of the manual annotations (Section 3) as follows. If a clue instance is not in a subjective expression (and therefore not in a sentiment expression), its gold class is neutral. If a clue instance appears in just one subjective expression or in multiple subjective expressions with the same contextual polarity, its gold class is the contextual polarity of the subjective expression(s). If a clue instance appears in a mixture of negative and neutral subjective expressions, its gold class is negative; if it is in a mixture of positive and neutral subjective expressions, its gold class is positive. Finally, if a clue instance appears in at least one positive and one negative subjective expression (or in a subjective expression marked as both), then its gold class is both. A clue instance can appear in more than one subjective expression because in the MPQA annotation scheme, it is possible for direct subjective frames and expressive subjective elements frames to overlap.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Definition of the Gold Standard",
                "sec_num": "5."
            },
            {
                "text": "Before delving into the task of recognizing contextual polarity, an important question to address is how useful prior polarity alone is for identifying contextual polarity. To answer this question, we create a classifier that simply assumes the contextual polarity of a clue instance is the same as the clue's prior polarity. We explore this classifier's performance on a small amount of development data, which is not part of the data used in the subsequent experiments.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "A Prior-Polarity Classifier",
                "sec_num": "6."
            },
            {
                "text": "This simple classifier has an accuracy of 48%. From the confusion matrix given in Table 6 , we see that 76% of the errors result from words with non-neutral prior polarity appearing in phrases with neutral contextual polarity. Only 12% of the errors result from words with neutral prior polarity appearing in expressions with non-neutral contextual polarity, and only 11% of the errors come from words with a positive or negative prior polarity appearing in expressions with the opposite contextual polarity. Table 6 also shows that positive clues tend to be used in negative expressions far more often than negative clues tend to be used in positive expressions.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 82,
                        "end": 89,
                        "text": "Table 6",
                        "ref_id": null
                    },
                    {
                        "start": 509,
                        "end": 516,
                        "text": "Table 6",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "A Prior-Polarity Classifier",
                "sec_num": "6."
            },
            {
                "text": "Given that by far the largest number of errors come from clues with positive, negative, or both prior polarity appearing in neutral contexts, we were motivated to try a two-step approach to the problem of sentiment classification. The first step, Neutral-Polar Classification, tries to determine if an instance is neutral or polar in context. The second step, Polarity Classification, takes all instances that step one classified as polar, and tries to disambiguate their contextual polarity. This two-step approach is illustrated in Figure 1 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 534,
                        "end": 542,
                        "text": "Figure 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "A Prior-Polarity Classifier",
                "sec_num": "6."
            },
            {
                "text": "The features used in our experiments were motivated both by the literature and by exploration of the contextual-polarity annotations in our development data. A number Table 6 Confusion matrix for the prior-polarity classifier on the development set. of features were inspired by the paper on contextual-polarity influencers by Polanyi and Zaenan (2004) . Other features are those that have been found useful in the past for recognizing subjective sentences (Wiebe, Bruce, and O'Hara 1999; Wiebe and Riloff 2005) .",
                "cite_spans": [
                    {
                        "start": 327,
                        "end": 352,
                        "text": "Polanyi and Zaenan (2004)",
                        "ref_id": null
                    },
                    {
                        "start": 457,
                        "end": 488,
                        "text": "(Wiebe, Bruce, and O'Hara 1999;",
                        "ref_id": "BIBREF38"
                    },
                    {
                        "start": 489,
                        "end": 511,
                        "text": "Wiebe and Riloff 2005)",
                        "ref_id": "BIBREF40"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 167,
                        "end": 174,
                        "text": "Table 6",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Features",
                "sec_num": "7."
            },
            {
                "text": "For distinguishing between neutral and polar instances, we use the features listed in Table 7 . For ease of description, we group the features into six sets: word features, general modification features, polarity modification features, structure features, sentence features, and one document feature. Word Features In addition to the word token (the token of the clue instance being classified), the word features include the parts of speech of the previous word, the word itself, and the next word. The prior polarity and reliability class features represent those pieces of information about the clue which are taken from the lexicon.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 86,
                        "end": 93,
                        "text": "Table 7",
                        "ref_id": "TABREF5"
                    }
                ],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "General Modification Features These are binary features that capture different types of relationships involving the clue instance.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "The first four features involve relationships with the word immediately before or after the clue instance. The preceded by adjective feature is true if the clue instance is a noun preceded by an adjective. The preceded by adverb feature is true if the preceding word is an adverb other than not. The preceded by intensifier feature is true if the preceding word is an intensifier, and the self intensifier feature is true if the clue instance itself is an intensifier. A word is considered to be an intensifier if it appears in a list of intensifiers and if it precedes a word of the appropriate part of speech (e.g., an intensifier adjective must come before a noun). The list of intensifiers is a compilation of those listed in Quirk et al. (1985) , intensifiers identified from existing entries in the subjectivity lexicon, and intensifiers identified during explorations of the development data.",
                "cite_spans": [
                    {
                        "start": 730,
                        "end": 749,
                        "text": "Quirk et al. (1985)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "The modifies/modifed by features involve the dependency parse tree of the sentence, obtained by first parsing the sentence (Collins 1997) and then converting the tree into its dependency representation (Xia and Palmer 2001) . In a dependency representation, every node in the tree structure is a surface word (i.e., there are no abstract nodes such as NP or VP). The parent word is called the head, and its children are its modifiers. The Sentence Features strongsubj clues in current sentence: 0, 1, 2, 3 (or more) strongsubj clues in previous sentence: 0, 1, 2, 3 (or more) strongsubj clues in next sentence: 0, 1, 2, 3 (or more) weaksubj clues in current sentence: 0, 1, 2, 3 (or more) weaksubj clues in previous sentence: 0, 1, 2, 3 (or more) weaksubj clues in next sentence: 0, 1, 2, 3 (or more) adjectives in sentence: 0, 1, 2, 3 (or more) adverbs in sentence (other than not): 0, 1, 2, 3 (or more) cardinal number in sentence: binary pronoun in sentence: binary modal in sentence (other than will): binary Document Feature document topic/domain edge between a parent and a child specifies the grammatical relationship between the two words. Figure 2 shows an example of a dependency parse tree. Instances of clues in the tree are marked with the clue's prior polarity and reliability class from the lexicon.",
                "cite_spans": [
                    {
                        "start": 123,
                        "end": 137,
                        "text": "(Collins 1997)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 202,
                        "end": 223,
                        "text": "(Xia and Palmer 2001)",
                        "ref_id": "BIBREF43"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1148,
                        "end": 1156,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "For each clue instance, the modifies/modifed by features capture whether there are adj, mod, or vmod relationships between the clue instance and any other instances from the lexicon. Specifically, the modifies strongsubj feature is true if the clue instance and its parent share an adj, mod, or vmod relationship, and if its parent is an instance of a strongsubj clue from the lexicon. The modifies weaksubj feature is the same, except that it looks in the parent for an instance of a weaksubj clue. The modified by strongsubj The dependency tree for the sentence The human rights report poses a substantial challenge to the U.S. interpretation of good and evil. Prior polarity and reliability class are marked in parentheses for words that match clues from the lexicon.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "feature is true for a clue instance if one of its children is an instance of a strongsubj clue, and if the clue instance and its child share an adj, mod, or vmod relationship. The modified by weaksubj feature is the same, except that it looks for instances of weaksubj clues in the children. Although the adj and vmod relationships are typically local, the mod relationship involves longer-distance as well as local dependencies. Figure 2 helps to illustrate these features. The modifies weaksubj feature is true for substantial, because substantial modifies challenge, which is an instance of a weaksubj clue. For rights, the modifies weaksubj feature is false, because rights modifies report, which is not an instance of a weaksubj clue. The modified by weaksubj feature is false for substantial, because it has no modifiers that are instances of weaksubj clues. For challenge, the modified by weaksubj feature is true because it is being modified by substantial, which is an instance of a weaksubj clue.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 430,
                        "end": 438,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "Polarity Modification Features The modifies polarity, modified by polarity, and conj polarity features capture specific relationships between the clue instance and other sentiment clues it may be related to. If the clue instance and its parent in the dependency tree share an obj, adj, mod, or vmod relationship, the modifies polarity feature is set to the prior polarity of the parent. If the parent is not in the prior-polarity lexicon, its prior polarity is considered neutral. If the clue instance is at the root of the tree and has no parent, the value of the feature is notmod. The modified by polarity feature is similar, looking for adj, mod, and vmod relationships and other sentiment clues in the children of the clue instance. The conj polarity feature determines if the clue instance is in a conjunction. If so, the value of this feature is its sibling's prior polarity. As before, if the sibling is not in the lexicon, its prior polarity is neutral. If the clue instance is not in a conjunction, the value for this feature is notmod. Figure 2 also helps to illustrate these modification features. The word substantial with positive prior polarity modifies the word challenge with negative prior polarity. Therefore the modifies polarity feature is negative for substantial, and the modified by polarity feature is positive for challenge. The words good and evil are in a conjunction together; thus the conj polarity feature is negative for good and positive for evil.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 1047,
                        "end": 1055,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "Structure Features These are binary features that are determined by starting with the clue instance and climbing up the dependency parse tree toward the root, looking for particular relationships, words, or patterns. The in subject feature is true if we find a subj relationship on the path to the root. The in copular feature is true if in subject is false and if a node along the path is both a main verb and a copular verb. The in passive feature is true if a passive verb pattern is found on the climb.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "The in subject and in copular features were motivated by the intuition that the syntactic role of a word may influence whether a word is being used to express a sentiment. For example, consider the word polluters in each of the following two sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "(12) Under the application shield, polluters are allowed to operate if they have a permit.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "(13) \"The big-city folks are pointing at the farmers and saying you are polluters . . . \"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "In the first sentence, polluters is simply being used as a referring expression. In the second sentence, polluters is clearly being used to express a negative evaluation of the farmers. The motivation for the in passive feature was previous work by Riloff and Wiebe (2003) , who found that different words are more or less likely to be subjective depending on whether they are in the active or passive. Sentence Features These are features that previously were found useful for sentence-level subjectivity classification (Wiebe, Bruce, and O'Hara 1999; Wiebe and Riloff 2005) . They include counts of strongsubj and weaksubj clue instances in the current, previous and next sentences, counts of adjectives and adverbs other than not in the current sentence, and binary features to indicate whether the sentence contains a pronoun, a cardinal number, and a modal other than will. Document Feature There is one document feature representing the topic or domain of the document. The motivation for this feature is that whether or not a word is expressing a sentiment or even a private state in general may depend on the subject of the discourse. For example, the words fever and sufferer may express a negative sentiment in certain contexts, but probably not in a health or medical context, as is the case in the following sentence.",
                "cite_spans": [
                    {
                        "start": 249,
                        "end": 272,
                        "text": "Riloff and Wiebe (2003)",
                        "ref_id": "BIBREF26"
                    },
                    {
                        "start": 521,
                        "end": 552,
                        "text": "(Wiebe, Bruce, and O'Hara 1999;",
                        "ref_id": "BIBREF38"
                    },
                    {
                        "start": 553,
                        "end": 575,
                        "text": "Wiebe and Riloff 2005)",
                        "ref_id": "BIBREF40"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "(14) The disease can be contracted if a person is bitten by a certain tick or if a person comes into contact with the blood of a congo fever sufferer.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "In the creation of the MPQA corpus, about two-thirds of the documents were selected to be on one of the 10 topics listed in Table 8 . The documents for each topic were identified by human searches and by an information retrieval system. The remaining documents were semi-randomly selected from a very large pool of documents from the world press. In the corpus, these documents are listed with the topic miscellaneous. Rather than leaving these documents unlabeled, we chose to label them using the following general domain categories: economics, general politics, health, report events, and war and terrorism. Table 9 lists the features that we use for step two, polarity classification. Word token, word prior polarity, and the polarity-modification features are the same as described for neutral-polar classification. We use two features to capture two different types of negation. The negated feature is a binary feature that is used to capture more local negations: Its value is true if a negation word or phrase is found within the four words preceding the clue instance, and if the negation word is not also in a phrase that acts as an intensifier rather than a negator. Examples of phrases that intensify rather than negate are not only and nothing if not. The negated subject feature captures a longer-distance type of negation. This feature is true if the subject of the clause containing the clue instance is negated. For example, the negated subject feature is true for support in the following sentence.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 124,
                        "end": 131,
                        "text": "Table 8",
                        "ref_id": "TABREF6"
                    },
                    {
                        "start": 611,
                        "end": 618,
                        "text": "Table 9",
                        "ref_id": "TABREF7"
                    }
                ],
                "eq_spans": [],
                "section": "Features for Neutral-Polar Classification",
                "sec_num": "7.1"
            },
            {
                "text": "(15) No politically prudent Israeli could support either of them.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Polarity Classification",
                "sec_num": "7.2"
            },
            {
                "text": "The last three polarity features look in a window of four words before the clue instance, searching for the presence of particular types of polarity influencers. General polarity shifters reverse polarity (e.g., little truth, little threat). Negative polarity shifters typically make the polarity of an expression negative (e.g., lack of understanding). Positive polarity shifters typically make the polarity of an expression positive (e.g., abate the damage). The polarity influencers that we used were identified through explorations of the development data.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Features for Polarity Classification",
                "sec_num": "7.2"
            },
            {
                "text": "We have two primary goals with our experiments in recognizing contextual polarity. The first is to evaluate the features described in Section 7 as to their usefulness for this task. The second is to investigate the importance of recognizing neutral instancesrecognizing when a sentiment clue is not being used to express a sentiment-for classifying contextual polarity.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "To evaluate features, we investigate their performance, both together and separately, across several different learning algorithms. Varying the learning algorithm allows us to verify that the features are robust and that their performance is not the artifact of a particular algorithm. We experiment with four different types of machine learning: boosting, memory-based learning, rule learning, and support vector learning. For boosting, we use BoosTexter (Schapire and Singer 2000) AdaBoost.MH. For rule learning, we use Ripper (Cohen 1996) . For memory-based learning, we use TiMBL (Daelemans et al. 2003b ) IB1 (k-nearest neighbor). For support vector learning, we use SVM-light and SVM-multiclass (Joachims 1999) . SVM-light is used for the experiments involving binary classification (neutral-polar classification), and SVM-multiclass is used for experiments with more than two classes. These machine learning algorithms were chosen because they have been used successfully for a number of natural language processing tasks, and they represent several different types of learning.",
                "cite_spans": [
                    {
                        "start": 529,
                        "end": 541,
                        "text": "(Cohen 1996)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 584,
                        "end": 607,
                        "text": "(Daelemans et al. 2003b",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 701,
                        "end": 716,
                        "text": "(Joachims 1999)",
                        "ref_id": "BIBREF16"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "For all of the classification algorithms except for SVM, the features for a clue instance are represented as they are presented in Section 7. For SVM, the representations for numeric and discrete-valued features are changed. Numeric features, such as the count of strongsubj clue instances in a sentence, are scaled to range between 0 and 1. Discrete-valued features, such as the reliability class feature, are converted into multiple binary features. For example, the reliability class feature is represented by two binary features: one for whether the clue instance is a strongsubj clue and one for whether the clue instance is a weaksubj clue.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "To investigate the importance of recognizing neutral instances, we perform two sets of polarity classification (step two) experiments. First, we experiment with classifying the polarity of all gold-standard polar instances-the clue instances identified as polar in context by the manual polarity annotations. Second, we experiment with using the polar instances identified automatically by the neutral-polar classifiers. Because the second set of experiments includes the neutral instances misclassified in step one, we can compare results for the two sets of experiments to see how the noise of neutral instances affects the performance of the polarity features.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "All experiments are performed using 10-fold cross validation over a test set of 10,287 sentences from 494 MPQA corpus documents. We measure performance in terms of accuracy, recall, precision, and F-measure. Accuracy is simply the total number of instances correctly classified. Recall, precision, and F-measure for a given class C are defined as follows. Recall is the percentage of all instances of class C correctly identified.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "Rec(C) = | instances of C correctly identified | | all instances of C |",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "Precision is the percentage of instances classified as class C that are class C in truth.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "Prec(C) = | instances of C correctly identified | | all instances identified as C | F-measure is the harmonic mean of recall and precision.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "F(C) = 2 \u00d7Rec(C) \u00d7 Prec(C) Rec(C) + Prec(C)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "All results reported are averages over the 10 folds.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments in Recognizing Contextual Polarity",
                "sec_num": "8."
            },
            {
                "text": "In our two-step process for recognizing contextual polarity, the first step is neutral-polar classification, determining whether each instance of a clue from the lexicon is neutral or polar in context. In our test set, there are 26,729 instances of clues from the lexicon. The features we use for this step were listed above in Table 7 and described in Section 7.1. In this section, we perform two sets of experiments. In the first, we compare the results of neutral-polar classification using all the neutral-polar features against two baselines. The first baseline uses just the word token feature. The second baseline (word+priorpol) uses the word token and prior polarity features. In the second set of experiments, we explore the performance of different sets of features for neutral-polar classification.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 328,
                        "end": 335,
                        "text": "Table 7",
                        "ref_id": "TABREF5"
                    }
                ],
                "eq_spans": [],
                "section": "Neutral-Polar Classification",
                "sec_num": "8.1"
            },
            {
                "text": "Research has shown that the performance of learning algorithms for NLP tasks can vary widely depending on their parameter settings, and that the optimal parameter settings can also vary depending on the set of features being evaluated (Daelemans et al. 2003a; Hoste 2005) . Although the goal of this work is not to identify the optimal configuration for each algorithm and each set of features, we still want to make a reasonable attempt to find a good configuration for each algorithm. To do this, we perform 10-fold cross validation of the more challenging baseline classifier (word+priorpol) on the development data, varying select parameter settings. The results from those experiments are then used to select the parameter settings for each algorithm. For BoosTexter, we vary the number of rounds of boosting. For TiMBL, we vary the value for k (the number of neighbors) and the distance metric (overlap or modified value difference metric [MVDM] ). For Ripper, we vary whether negative tests are disallowed for nominal (-!n) and set (-!s) valued attributes and how much to simplify the hypothesis (-S). For SVM, we experiment with linear, polynomial, and radial basis function kernels. Table 10 gives the settings selected for the neutral-polar classification experiments for the different learning algorithms. Table 11 . For each algorithm, we give the results for the two baseline classifiers, followed by the results for the classifier trained using all the neutral-polar features. The results shown in bold are significantly better than both baselines (two-sided t-test, p \u2264 0.05) for the given algorithm. Working together, how well do the neutral-polar features perform? For BoosTexter, TiMBL, and Ripper, the classifiers trained using all the features improve significantly over the two baselines in terms of accuracy, polar recall, polar F-measure, and neutral precision. Neutral F-measure is also higher, but not significantly so. These consistent results across three of the four algorithms show that the neutral-polar features are helpful for determining when a sentiment clue is actually being used to express a sentiment.",
                "cite_spans": [
                    {
                        "start": 235,
                        "end": 259,
                        "text": "(Daelemans et al. 2003a;",
                        "ref_id": null
                    },
                    {
                        "start": 260,
                        "end": 271,
                        "text": "Hoste 2005)",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 945,
                        "end": 951,
                        "text": "[MVDM]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1192,
                        "end": 1200,
                        "text": "Table 10",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 1317,
                        "end": 1325,
                        "text": "Table 11",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Neutral-Polar Classification",
                "sec_num": "8.1"
            },
            {
                "text": "Interestingly, Ripper is the only algorithm for which the word-token baseline performed better than the word+priorpol baseline. Nevertheless, the prior polarity feature is an important component in the performance of the Ripper classifier using all the features. Excluding prior polarity from this classifier results in a significant decrease in Table 11 Results for neutral-polar classification (step one). performance for every metric. Decreases range from 2.5% for neutral recall to 9.5% for polar recall. The best SVM classifier is the word+priorpol baseline. In terms of accuracy, this classifier does not perform much worse than the BoosTexter and TiMBL classifiers that use all the neutral-polar features: The SVM word+priorpol baseline classifier has an accuracy of 75.6%, and both the BoosTexter and TiMBL classifiers have an accuracy of 76.5%. However, the BoosTexter and TiMBL classifiers using all the features perform notably better in terms of polar recall and F-measure. The BoosTexter and TiMBL classifiers have polar recalls that are 7% and 9.2% higher than the SVM baseline. Polar F-measures for BoosTexter and TiMBL are 3.9% and 4.5% higher. These increases are significant for p \u2264 0.01.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 346,
                        "end": 354,
                        "text": "Table 11",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Neutral-Polar Classification",
                "sec_num": "8.1"
            },
            {
                "text": "To evaluate the contribution of the various features for neutral-polar classification, we perform a series of experiments in which different sets of neutral-polar features are added to the word+priorpol baseline and new classifiers are trained. We then compare the performance of these new classifiers to the word+priorpol baseline, with the exception of the Ripper classifiers, which we compare to the higher word baseline. Table 12 lists the sets of features tested in these experiments. The feature sets generally correspond to how the neutral-polar features are presented in Table 7 , although some of the groups are broken down into more fine-grained sets that we believe capture meaningful distinctions. Table 13 gives the results for these experiments. Increases and decreases for a given metric as compared to the word+priorpol baseline (word baseline for Ripper) are indicated by + or -, respectively. Where changes are significant at the p \u2264 0.1 level, ++ or --are used, and where changes are significant at the p \u2264 0.05 level, +++ or --are used. An \"nc\" indicates no change (a change of less than \u00b1 0.05) compared to the baseline.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 425,
                        "end": 433,
                        "text": "Table 12",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 579,
                        "end": 586,
                        "text": "Table 7",
                        "ref_id": "TABREF5"
                    },
                    {
                        "start": 710,
                        "end": 718,
                        "text": "Table 13",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Feature Set Evaluation.",
                "sec_num": "8.1.2"
            },
            {
                "text": "What does Table 13 reveal about the performance of various feature sets for neutralpolar classification? Most noticeable is that no individual feature sets stand out as strong performers. The only significant improvements in accuracy come from the PARTS-OF-SPEECH and RELIABILITY-CLASS feature sets for Ripper. These improvements are perhaps not surprising given that the Ripper baseline was much lower to begin with. Very few feature sets show any improvement for SVM. Again, this is not unexpected given that all the features together performed worse than the word+priorpol baseline for SVM. The performance of the feature sets for BoosTexter and TiMBL are perhaps the most revealing. In the previous experiments using all the features together, these algorithms produced classifiers with the same high performance. In these experiments, six different feature sets for each algorithm show improvements in accuracy over the baseline, yet none of those improvements are significant. This suggests that achieving the highest performance for neutral-polar classification requires a wide variety of features working together in combination. We further test this result by evaluating the effect of removing the features that produced either no change or a drop in accuracy from the respective all-feature classifiers. For example, we train a TiMBL neutral-polar classifier using all the features except for those in the PRECEDED-POS, INTENSIFY, STRUCTURE, CURSENT-COUNTS, and TOPIC feature sets, and then compare the performance of this new classifier to the TiMBL, allfeature classifier. Although removing the non-performing features has little effect for BoosTexter, performance does drop for both TiMBL and Ripper. The primary source of this performance drop is a decrease in polar recall: 2% for TiMBL and 3.2% for Ripper.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 10,
                        "end": 18,
                        "text": "Table 13",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Feature Set Evaluation.",
                "sec_num": "8.1.2"
            },
            {
                "text": "Although no feature sets stand out in Table 13 as far as giving an overall high performance, there are some features that consistently improve performance across the different algorithms. The reliability class of the clue instance (RELIABILITY-CLASS) improves accuracy over the baseline for all four algorithms. It is the only feature that does so. The RELCLASS-MOD features give improvements for all metrics for BoosTexter, Ripper, and TiMBL, as well as improving polar F-measure for SVM. The PARTS-OF-SPEECH features are also fairly consistent, improving performance for all the algorithms except for SVM. There are also a couple of feature sets that consistently do not improve performance for any of the algorithms: the INTENSIFY and PRECEDED-POS features.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 38,
                        "end": 46,
                        "text": "Table 13",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Feature Set Evaluation.",
                "sec_num": "8.1.2"
            },
            {
                "text": "For the second step of recognizing contextual polarity, we classify the polarity of all clue instances identified as polar in step one. The features for polarity classification were listed in Table 9 and described in Section 7.2.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 192,
                        "end": 199,
                        "text": "Table 9",
                        "ref_id": "TABREF7"
                    }
                ],
                "eq_spans": [],
                "section": "Polarity Classification",
                "sec_num": "8.2"
            },
            {
                "text": "We investigate the performance of the polarity features under two conditions: (1) perfect neutral-polar recognition and (2) automatic neutral-polar recognition. For condition 1, we identify the polar instances according to the gold-standard, manual contextual-polarity annotations. In the test data, 9,835 instances of the clues from the lexicon are polar in context according to the manual annotations. Experiments under condition 1 classify these instances as having positive, negative, or both (positive and negative) polarity. For condition 2, we take the best performing neutral-polar classifier for each algorithm and use the output from those algorithms to identify the polar instances. Because polar instances now are being identified automatically, there will be noise in the form of misclassified neutral instances. Therefore, for experiments under condition 2 we include the neutral class and perform four-way classification instead of three-way. Condition 1 allows us to investigate the performance of the different polarity features without the noise of misclassified neutral instances. Also, because the set of polar instances being classified is the same for all the algorithms, condition 1 allows us to compare the performance of the polarity features across the different algorithms. However, condition 2 is the more natural one. It allows us to see how the noise of neutral instances affects the performance of the polarity features.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Polarity Classification",
                "sec_num": "8.2"
            },
            {
                "text": "The following sections describe three sets of experiments. First, we investigate the performance of the polarity features used together for polarity classification under condition 1. As before, the word and word+priorpol classifiers provide our baselines. In the second set of experiments, we explore the performance of different sets of features for polarity classification, again assuming perfect recognition of the polar instances. Finally, we experiment with polarity classification using all the polarity features under condition 2, automatic recognition of the polar instances.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Polarity Classification",
                "sec_num": "8.2"
            },
            {
                "text": "As before, we use the development data to select the parameter settings for each algorithm. The settings for polarity classification are given in Table 14 . They were selected based on the performance of the word+priorpol baseline classifier under condition 2.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 146,
                        "end": 154,
                        "text": "Table 14",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Polarity Classification",
                "sec_num": "8.2"
            },
            {
                "text": ": Condition 1. The results for polarity classification using all the polarity features, assuming perfect neutral-polar recognition for step one, are given in Table 15 . For each algorithm, we give the results for the two baseline classifiers, followed by the results for the classifier trained using all the polarity features. For the metrics where the polarity features perform statistically better than both baselines (two-sided t-test, p \u2264 0.05), the results are given in bold.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 158,
                        "end": 166,
                        "text": "Table 15",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Classification Results",
                "sec_num": "8.2.1"
            },
            {
                "text": "Algorithm settings for polarity classification.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Table 14",
                "sec_num": null
            },
            {
                "text": "BoosTexter 2,000 rounds of boosting TiMBL k=1, MVDM distance metric Ripper -!s, -S 0.5 SVM linear kernel Table 15 Results for polarity classification (step two) using gold-standard polar instances.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 105,
                        "end": 113,
                        "text": "Table 15",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Algorithm Settings",
                "sec_num": null
            },
            {
                "text": "Acc How well do the polarity features perform working all together? For all algorithms, the polarity classifier using all the features significantly outperforms both baselines in terms of accuracy, positive F-measure, and negative F-measure. These consistent improvements in performance across all four algorithms show that these features are quite useful for polarity classification.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Positive Negative Both",
                "sec_num": null
            },
            {
                "text": "One interesting thing that Table 15 reveals is that negative polarity words are much more straightforward to recognize than positive polarity words, at least in this corpus. For the negative class, precisions and recalls for the word+priorpol baseline range from 82.2 to 87.2. For the positive class, precisions and recalls for the word+priorpol baseline range from 63.7 to 76.7. However, it is with the positive class that polarity features seem to help the most. With the addition of the polarity features, positive F-measure improves by 5 points on average; improvements in negative F-measures average only 2.75 points.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 27,
                        "end": 35,
                        "text": "Table 15",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Positive Negative Both",
                "sec_num": null
            },
            {
                "text": "To evaluate the performance of the various features for polarity classification, we again perform a series of ablation experiments. As before, we start with the word+priorpol baseline classifier, add different sets of polarity features, train new classifiers, and compare the results of the new classifiers to the baseline. Increases and decreases for a given metric as compared to the word+priorpol baseline are indicated by + or -, respectively; ++ or --indicates the change is significant at the p < 0.1 level; +++ or ---indicates significance at the p < 0.05 level. Table 16 lists the sets of features tested in each experiment, and Table 17 shows the results of the experiments. Results are reported as they were previously in Section 8.1.2, with increases and decreases compared to the baseline for a given metric indicated by + or -, respectively. Looking at Table 17 , we see that all three sets of polarity features help to increase performance as measured by accuracy and positive and negative F-measures. This is true for all the classification algorithms. As we might expect, including the negation features has the most marked effect on the performance of polarity classification, with statistically significant improvements for most metrics across all the algorithms. 9 The polarity-modification features also seem to be important for polarity classification, in particular for disambiguating the positive instances. For all the algorithms except TiMBL, including the polarity-modification features results in significant improvements for at least one of the positive metrics. The polarity shifters also help classification, but they seem to be the weakest of the features: Including them does not result in significant improvements for any algorithm.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 570,
                        "end": 578,
                        "text": "Table 16",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 637,
                        "end": 645,
                        "text": "Table 17",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 866,
                        "end": 874,
                        "text": "Table 17",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Feature Set Evaluation.",
                "sec_num": "8.2.2"
            },
            {
                "text": "Another question that is interesting to consider is how much the word token feature contributes to polarity classification, given all the other polarity features. Is it enough to know the prior polarity of a word, whether it is being negated, and how it is related to other polarity influencers? To answer this question, we train classifiers using all the polarity features except for word token. Table 18 gives the results for these classifiers; for comparison, the results for the all-feature polarity classifiers are also given. Interestingly, excluding the word token feature produces only small changes in the overall results. The results for BoosTexter and Ripper are slightly lower, and the results for SVM are practically unchanged. TiMBL actually shows a slight improvement, with the exception of the both class. This provides further evidence of the strength of the polarity features. Also, a classifier not tied to actual word tokens may potentially be a more domain-independent classifier.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 397,
                        "end": 405,
                        "text": "Table 18",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Feature Set Evaluation.",
                "sec_num": "8.2.2"
            },
            {
                "text": "The experiments in Section 8.2.1 show that the polarity features perform well under the ideal condition of perfect recognition of polar instances. The next question to consider is how well the polarity features perform under the more natural but less-than-perfect condition of automatic recognition of polar instances. To investigate this, the polarity classifiers (including the baselines) for each algorithm in these experiments start with the polar instances identified by the best performing neutral-polar classifier for that algorithm (from Section 8.1.1). The results for these experiments are given in Table 19 . As before, statistically significant improvements over both baselines are given in bold.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 609,
                        "end": 617,
                        "text": "Table 19",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Classification Results: Condition 2.",
                "sec_num": "8.2.3"
            },
            {
                "text": "How well do the polarity features perform in the presence of noise from misclassified neutral instances? Our first observation comes from comparing Table 15 with  Table 19 : Polarity classification results are much lower for all classifiers with the noise of neutral instances. Yet in spite of this, the polarity features still produce classifiers that ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 148,
                        "end": 171,
                        "text": "Table 15 with  Table 19",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Classification Results: Condition 2.",
                "sec_num": "8.2.3"
            },
            {
                "text": "Results for polarity classification (step two) using automatically identified polar instances. outperform the baselines. For three of the four algorithms, the classifier using all the polarity features has the highest accuracy. For BoosTexter and TiMBL, the improvements in accuracy over both baselines are significant. Also for all algorithms, using the polarity features gives the highest positive and negative F-measures. Because the set of polarity instances being classified by each algorithm is different, we cannot directly compare the results from one algorithm to the next.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Table 19",
                "sec_num": null
            },
            {
                "text": "Although the two-step approach to recognizing contextual polarity allows us to focus our investigation on the performance of features for both neutral-polar classification and polarity classification, the question remains: How does the two-step approach compare to recognizing contextual polarity in a single classification step? The results shown in Table 20 help to answer this question. The first row in Table 20 for each algorithm shows the combined result for the two stages of classification. For BoosTexter, TiMBL, and Ripper, this is the combination of results from using all the neutral-polar features for step one, together with the results from using all of the polarity features for step two. 10 For SVM, this is the combination of results from the word+priorpol baseline from step one, together with results for using all the polarity features for step two. Recall that the word+priorpol classifier was the best neutral-polar classifier for SVM (see Table 11 ). The second rows for BoosTexter, TiMBL, and Ripper show the results of a single classifier trained to recognize contextual polarity using all the neutral-polar and polarity features together. For SVM, the second row shows the results of classifying the contextual polarity using just the word token feature. This classifier outperformed all others for SVM. In the table, the best result for each metric for each algorithm is highlighted in bold.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 351,
                        "end": 359,
                        "text": "Table 20",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 407,
                        "end": 415,
                        "text": "Table 20",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 963,
                        "end": 971,
                        "text": "Table 11",
                        "ref_id": "TABREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Two-step versus One-step Recognition of Contextual Polarity",
                "sec_num": "8.3"
            },
            {
                "text": "When comparing the two-step and one-step approaches, contrary to our expectations, we see that the one-step approach performs about as well or better than the two-step approach for recognizing contextual polarity. For SVM, the improvement in accuracy achieved by the two-step approach is significant, but this is not true for the other algorithms. One fairly consistent difference between the two approaches is that the two-step approach scores slightly higher for neutral F-measure, and the onestep approach achieves higher F-measures for the polarity classes. The difference in negative F-measure is significant for BoosTexter, TiMBL, and Ripper. The exception to this is SVM. For SVM, the two-step approach achieves significantly higher positive and negative F-measures.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Two-step versus One-step Recognition of Contextual Polarity",
                "sec_num": "8.3"
            },
            {
                "text": "One last question we consider is how much the neutral-polar features contribute to the performance of the one-step classifiers. The third line in Table 20 for BoosTexter, TiMBL, and Ripper gives the results for a one-step classifier trained without the neutralpolar features. Although the differences are not always large, excluding the neutralpolar features consistently degrades performance in terms of accuracy and positive, negative, and neutral F-measures. The drop in negative F-measure is significant for all three algorithms, the drop in neutral F-measure is significant for BoosTexter and TiMBL, and the drop in accuracy is significant for TiMBL and Ripper (and for BoosTexter at the p \u2264 0.1 level).",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 146,
                        "end": 154,
                        "text": "Table 20",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Two-step versus One-step Recognition of Contextual Polarity",
                "sec_num": "8.3"
            },
            {
                "text": "Results for contextual polarity classification for both two-step and one-step approaches. The modest drop in performance that we see when excluding the neutral-polar features in the one-step approach seems to suggest that discriminating between neutral and polar instances is helpful but not necessarily crucial. However, consider Figure 3 . In this figure, we show the F-measures for the positive, negative, and both classes for the BoosTexter polarity classifier that uses the gold-standard neutral/polar instances (from Table 15 ) and for the BoosTexter one-step polarity classifier that uses all features (from Table 20 ). Plotting the same sets of results for the other three algorithms produces very similar figures. The difference when the classifiers have to contend with the noise from neutral instances is dramatic. Although Table 20 shows that there is room for improvement across all the contextual polarity classes, Figure 3 shows us that perhaps the best way to achieve these improvements is to improve the ability to discriminate the neutral class from the others.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 331,
                        "end": 339,
                        "text": "Figure 3",
                        "ref_id": null
                    },
                    {
                        "start": 523,
                        "end": 531,
                        "text": "Table 15",
                        "ref_id": "TABREF0"
                    },
                    {
                        "start": 615,
                        "end": 623,
                        "text": "Table 20",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 835,
                        "end": 843,
                        "text": "Table 20",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 929,
                        "end": 937,
                        "text": "Figure 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Table 20",
                "sec_num": null
            },
            {
                "text": "Chart showing the positive, negative, and both class F-measures for the BoosTexter classifier that uses the gold-standard neutral/polar classes and the BoosTexter one-step classifier that uses all the features.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Figure 3",
                "sec_num": null
            },
            {
                "text": "Other researchers who have worked on classifying the contextual polarity of sentiment expressions are Yi et al. (2003) , Popescu and Etzioni (2005) , and Suzuki, Takamura, and Okumura (2006) . Yi et al. use a lexicon and manually developed patterns to classify contextual polarity. Their patterns are high-quality, yielding quite high precision over the set of expressions that they evaluate. Popescu and Etzioni use an unsupervised classification technique called relaxation labeling (Hummel and Zucker 1983) to recognize the contextual polarity of words that are at the heads of select opinion phrases. They take an iterative approach, using relaxation labeling first to determine the contextual polarities of the words, then again to label the polarities of the words with respect to their targets. A third stage of relaxation labeling then is used to assign final polarities to the words, taking into consideration the presence of other polarity terms and negation. As we do, Popescu and Etzioni use features that represent conjunctions and dependency relations between polarity words. Suzuki et al. use a bootstrapping approach to classify the polarity of tuples of adjectives and their target nouns in Japanese blogs. Included in the features that they use are the words that modify the adjectives and the word that the adjective modifies. They consider the effect of a single negation term, the Japanese equivalent of not.",
                "cite_spans": [
                    {
                        "start": 102,
                        "end": 118,
                        "text": "Yi et al. (2003)",
                        "ref_id": "BIBREF44"
                    },
                    {
                        "start": 121,
                        "end": 147,
                        "text": "Popescu and Etzioni (2005)",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 150,
                        "end": 190,
                        "text": "and Suzuki, Takamura, and Okumura (2006)",
                        "ref_id": "BIBREF31"
                    },
                    {
                        "start": 485,
                        "end": 509,
                        "text": "(Hummel and Zucker 1983)",
                        "ref_id": "BIBREF15"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrase-Level Sentiment Analysis",
                "sec_num": "9.1"
            },
            {
                "text": "Our work in recognizing contextual polarity differs from this research on expression-level sentiment analysis in several ways. First, the set of expressions they evaluate is limited either to those that target specific items of interest, such as products and product features, or to tuples of adjectives and nouns. In contrast, we seek to classify the contextual polarity of all instances of words from a large lexicon of subjectivity clues that appear in the corpus. Included in the lexicon are not only adjectives, but nouns, verbs, adverbs, and even modals.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrase-Level Sentiment Analysis",
                "sec_num": "9.1"
            },
            {
                "text": "Our work also differs from other research in the variety of features that we use. As other researchers do, we consider negation and the words that directly modify or are modified by the expression being classified. However, with negation, we have features for both local and longer-distance types of negation, and we take care to count negation terms only when they are actually being used to negate, excluding, for example, negation terms when they are used in phrases that intensify (e.g., not only). We also include contextual features to capture the presence of other clue instances in the surrounding sentences, and features that represent the reliability of clues from the lexicon.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrase-Level Sentiment Analysis",
                "sec_num": "9.1"
            },
            {
                "text": "Finally, a unique aspect of the work presented in this article is the evaluation of different features for recognizing contextual polarity. We first presented the features explored in this research in Wilson, Wiebe, and Hoffman (2005) , but this work significantly extends that initial evaluation. We explore the performance of features across different learning algorithms, and we evaluate not only features for discriminating between positive and negative polarity, but features for determining when a word is or is not expressing a sentiment in the first place (neutral in context). This is also the first work to evaluate the effect of neutral instances on the performance of features for discriminating between positive and negative contextual polarity.",
                "cite_spans": [
                    {
                        "start": 201,
                        "end": 234,
                        "text": "Wilson, Wiebe, and Hoffman (2005)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Phrase-Level Sentiment Analysis",
                "sec_num": "9.1"
            },
            {
                "text": "Recognizing contextual polarity is just one facet of the research in automatic sentiment analysis. Research ranges from work on learning the prior polarity (semantic orientation) of words and phrases (e.g., Hatzivassiloglou and McKeown 1997; Kamps and Marx 2002; Turney and Littman 2003; Hu and Liu 2004; Kim and Hovy 2004; Esuli and Sebastiani 2005; Takamura, Inui, and Okumura 2005; Popescu and Etzioni 2005; Andreevskaia and Bergler 2006; Esuli and Sebastiani 2006a; Kanayama and Nasukawa 2006) to characterizing the sentiment of documents, such as recognizing inflammatory messages (Spertus 1997) , tracking sentiment over time in online discussions (Tong 2001) , and classifying the sentiment of online messages (e.g., Das and Chen 2001; Koppel and Schler 2006) , customer feedback data (Gamon 2004) , or product and movie reviews (e.g., Turney 2002; Pang, Lee, and Vaithyanathan 2002; Dave, Lawrence, and Pennock 2003; Beineke, Hastie, and Vaithyanathan 2004; Mullen and Collier 2004; Bai, Padman, and Airoldi 2005; Whitelaw, Garg, and Argamon 2005; Kennedy and Inkpen 2006; Koppel and Schler 2006) .",
                "cite_spans": [
                    {
                        "start": 207,
                        "end": 241,
                        "text": "Hatzivassiloglou and McKeown 1997;",
                        "ref_id": null
                    },
                    {
                        "start": 242,
                        "end": 262,
                        "text": "Kamps and Marx 2002;",
                        "ref_id": null
                    },
                    {
                        "start": 263,
                        "end": 287,
                        "text": "Turney and Littman 2003;",
                        "ref_id": "BIBREF35"
                    },
                    {
                        "start": 288,
                        "end": 304,
                        "text": "Hu and Liu 2004;",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 305,
                        "end": 323,
                        "text": "Kim and Hovy 2004;",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 324,
                        "end": 350,
                        "text": "Esuli and Sebastiani 2005;",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 351,
                        "end": 384,
                        "text": "Takamura, Inui, and Okumura 2005;",
                        "ref_id": "BIBREF32"
                    },
                    {
                        "start": 385,
                        "end": 410,
                        "text": "Popescu and Etzioni 2005;",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 411,
                        "end": 441,
                        "text": "Andreevskaia and Bergler 2006;",
                        "ref_id": "BIBREF0"
                    },
                    {
                        "start": 442,
                        "end": 469,
                        "text": "Esuli and Sebastiani 2006a;",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 470,
                        "end": 497,
                        "text": "Kanayama and Nasukawa 2006)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 586,
                        "end": 600,
                        "text": "(Spertus 1997)",
                        "ref_id": "BIBREF28"
                    },
                    {
                        "start": 654,
                        "end": 665,
                        "text": "(Tong 2001)",
                        "ref_id": "BIBREF33"
                    },
                    {
                        "start": 724,
                        "end": 742,
                        "text": "Das and Chen 2001;",
                        "ref_id": "BIBREF6"
                    },
                    {
                        "start": 743,
                        "end": 766,
                        "text": "Koppel and Schler 2006)",
                        "ref_id": "BIBREF21"
                    },
                    {
                        "start": 792,
                        "end": 804,
                        "text": "(Gamon 2004)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 843,
                        "end": 855,
                        "text": "Turney 2002;",
                        "ref_id": "BIBREF34"
                    },
                    {
                        "start": 856,
                        "end": 890,
                        "text": "Pang, Lee, and Vaithyanathan 2002;",
                        "ref_id": null
                    },
                    {
                        "start": 891,
                        "end": 924,
                        "text": "Dave, Lawrence, and Pennock 2003;",
                        "ref_id": "BIBREF7"
                    },
                    {
                        "start": 925,
                        "end": 965,
                        "text": "Beineke, Hastie, and Vaithyanathan 2004;",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 966,
                        "end": 990,
                        "text": "Mullen and Collier 2004;",
                        "ref_id": "BIBREF24"
                    },
                    {
                        "start": 991,
                        "end": 1021,
                        "text": "Bai, Padman, and Airoldi 2005;",
                        "ref_id": null
                    },
                    {
                        "start": 1022,
                        "end": 1055,
                        "text": "Whitelaw, Garg, and Argamon 2005;",
                        "ref_id": "BIBREF36"
                    },
                    {
                        "start": 1056,
                        "end": 1080,
                        "text": "Kennedy and Inkpen 2006;",
                        "ref_id": "BIBREF19"
                    },
                    {
                        "start": 1081,
                        "end": 1104,
                        "text": "Koppel and Schler 2006)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Research in Sentiment Analysis",
                "sec_num": "9.2"
            },
            {
                "text": "Identifying prior polarity is a different task than recognizing contextual polarity, although the two tasks are complementary. The goal of identifying prior polarity is to automatically acquire the polarity of words or phrases for listing in a lexicon. Our work on recognizing contextual polarity begins with a lexicon of words with established prior polarities and then disambiguates in the corpus the polarity being expressed by the phrases in which instances of those words appear. To make the relationship between that task and ours clearer, some word lists that are used to evaluate methods for recognizing prior polarity (positive and negative word lists from the General Inquirer [Stone et al. 1966 ] and lists of positive and negative adjectives created for evaluation by Hatzivassiloglou and McKeown [1997] ) are included in the prior-polarity lexicon used in our experiments.",
                "cite_spans": [
                    {
                        "start": 687,
                        "end": 705,
                        "text": "[Stone et al. 1966",
                        "ref_id": "BIBREF29"
                    },
                    {
                        "start": 780,
                        "end": 815,
                        "text": "Hatzivassiloglou and McKeown [1997]",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Research in Sentiment Analysis",
                "sec_num": "9.2"
            },
            {
                "text": "For the most part, the features explored in this work differ from the ones used to identify prior polarity with just a few exceptions. Using a feature to capture conjunctions between clue instances was motivated in part by the work of Hatzivassiloglou and McKeown (1997) . They use constraints on the co-occurrence in conjunctions of words with similar or opposite polarity to predict the prior polarity of adjectives. Esuli and Sebastiani (2005) consider negation in some of their experiments involving WordNet glosses. Takamura et al. (2005) use negation words and phrases, including phrases such as lack of that are members in our lists of polarity shifters, and conjunctive expressions that they collect from corpora. Esuli and Sebastiani (2006a) is the only work in prior-polarity identification to include a neutral (objective) category and to consider a three-way classification between positive, negative, and neutral words. Although identifying prior polarity is a different task, they report a finding similar to ours, namely, that accuracy is lower when neutral words are included.",
                "cite_spans": [
                    {
                        "start": 235,
                        "end": 270,
                        "text": "Hatzivassiloglou and McKeown (1997)",
                        "ref_id": null
                    },
                    {
                        "start": 419,
                        "end": 446,
                        "text": "Esuli and Sebastiani (2005)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 521,
                        "end": 543,
                        "text": "Takamura et al. (2005)",
                        "ref_id": "BIBREF32"
                    },
                    {
                        "start": 722,
                        "end": 750,
                        "text": "Esuli and Sebastiani (2006a)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Research in Sentiment Analysis",
                "sec_num": "9.2"
            },
            {
                "text": "Some research in sentiment analysis classifies the sentiments of sentences. Morinaga et al. (2002) , Yu and Hatzivassiloglou (2003) , Kim and Hovy (2004) , Hu and Liu (2004) , and Grefenstette et al. (2004) 11 all begin by first creating prior-polarity lexicons. Yu and Hatzivassiloglou then assign a sentiment to a sentence by averaging the prior semantic orientations of instances of lexicon words in the sentence. Thus, they do not identify the contextual polarity of individual phrases containing clue instances, which is the focus of this work. Morinaga et al. only consider the positive or negative clue instance in each sentence that is closest to some target reference; Kim and Hovy, Hu and Liu, and Grefenstette et al. multiply or count the prior polarities of clue instances in the sentence.",
                "cite_spans": [
                    {
                        "start": 76,
                        "end": 98,
                        "text": "Morinaga et al. (2002)",
                        "ref_id": "BIBREF23"
                    },
                    {
                        "start": 101,
                        "end": 131,
                        "text": "Yu and Hatzivassiloglou (2003)",
                        "ref_id": "BIBREF45"
                    },
                    {
                        "start": 134,
                        "end": 153,
                        "text": "Kim and Hovy (2004)",
                        "ref_id": "BIBREF20"
                    },
                    {
                        "start": 156,
                        "end": 173,
                        "text": "Hu and Liu (2004)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 180,
                        "end": 206,
                        "text": "Grefenstette et al. (2004)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 678,
                        "end": 736,
                        "text": "Kim and Hovy, Hu and Liu, and Grefenstette et al. multiply",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Research in Sentiment Analysis",
                "sec_num": "9.2"
            },
            {
                "text": "These researchers also consider local negation to reverse polarity, with Morinaga et al. also taking into account the negating effect of words like insufficient. However, they do not use the other types of features that we consider in our experiments. Kaji and Kitsuregawa (2006) take a different approach to recognizing positive and negative sentences. They bootstrap from information easily obtained in \"Pro\" and \"Con\" HTML tables and lists, and from one high-precision linguistic pattern, to automatically construct a large corpus of positive and negative sentences. They then use this corpus to train a naive Bayes sentence classifier. In contrast to our work, sentiment classification in all of this research is restricted to identifying only positive and negative sentences (excluding our both and neutral categories). In addition, only one sentiment is assigned per sentence; our system assigns contextual polarity to individual expressions, which would allow for a sentence to be assigned to multiple sentiment categories. As we saw when exploring the contextual polarity annotations, it is not uncommon for sentences to contain more than one sentiment expression.",
                "cite_spans": [
                    {
                        "start": 252,
                        "end": 279,
                        "text": "Kaji and Kitsuregawa (2006)",
                        "ref_id": "BIBREF17"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Research in Sentiment Analysis",
                "sec_num": "9.2"
            },
            {
                "text": "Classifying the sentiment of documents is a very different task than recognizing the contextual polarity of words and phrases. However, some researchers have reported findings about document-level classification that are similar to our findings about phrase-level classification. Bai et al. (2005) argue that dependencies among key sentiment terms are important for classifying document sentiment. Similarly, we show that features for capturing when clue instances modify each other are important for phrase-level classification, in particular, for identifying positive expressions. Gamon (2004) achieves his best results for document classification using a wide variety of features, including rich linguistic features, such as features that capture constituent structure, features that combine part-of-speech and semantic relations (e.g., sentence subject or negated context), and features that capture tense information. We also achieve our best results for phrase-level classification using a wide variety of features, many of which are linguistically rich. Kennedy and Inkpen (2006) report consistently higher results for document sentiment classification when select polarity influencers, including negators and intensifiers, are included. 12 Koppel and Schler (2006) demonstrate the importance of neutral examples for document-level classification. In this work, we show that being able to correctly identify neutral instances is also very important for phraselevel sentiment analysis.",
                "cite_spans": [
                    {
                        "start": 280,
                        "end": 297,
                        "text": "Bai et al. (2005)",
                        "ref_id": null
                    },
                    {
                        "start": 583,
                        "end": 595,
                        "text": "Gamon (2004)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 1061,
                        "end": 1086,
                        "text": "Kennedy and Inkpen (2006)",
                        "ref_id": "BIBREF19"
                    },
                    {
                        "start": 1245,
                        "end": 1247,
                        "text": "12",
                        "ref_id": null
                    },
                    {
                        "start": 1248,
                        "end": 1272,
                        "text": "Koppel and Schler (2006)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Other Research in Sentiment Analysis",
                "sec_num": "9.2"
            },
            {
                "text": "Being able to determine automatically the contextual polarity of words and phrases is an important problem in sentiment analysis. In the research presented in this article, we tackle this problem and show that it is much more complex than simply determining whether a word or phrase is positive or negative. In our analysis of a corpus with annotations of subjective expressions and their contextual polarity, we find that positive and negative words from a lexicon are used in neutral contexts much more often than they are used in expressions of the opposite polarity. The importance of identifying when contextual polarity is neutral is further revealed in our classification experiments: When neutral instances are excluded, the performance of features for distinguishing between positive and negative polarity greatly improves.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions and Future Work",
                "sec_num": "10."
            },
            {
                "text": "A focus of this research is on understanding which features are important for recognizing contextual polarity. We experiment with a wide variety of linguistically motivated features, and we evaluate the performance of these features using several different machine learning algorithms. Features for distinguishing between neutral and polar instances are evaluated, as well as features for distinguishing between positive and negative contextual polarity. For classifying neutral and polar instances, we find that, although some features produce significant improvements over the baseline in terms of polar or neutral recall or precision, it is the combination of features together that is needed to achieve significant improvements in accuracy. For classifying positive and negative contextual polarity, features for capturing negation prove to be the most important. However, we find that features that also perform well are those that capture when a word is (or is not) modifying or being modified by other polarity terms. This suggests that identifying features that represent more complex interdependencies between polarity clues will be an important avenue for future research.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions and Future Work",
                "sec_num": "10."
            },
            {
                "text": "Another direction for future work will be to expand our lexicon using existing techniques for acquiring the prior polarity of words and phrases. It follows that a larger lexicon will have a greater coverage of sentiment expressions. However, expanding the lexicon with automatically acquired prior-polarity tags may result in an even greater proportion of neutral instances to contend with. Given the degradation in performance created by the neutral instances, whether expanding the lexicon automatically will result in improved performance for recognizing contextual polarity is an empirical question.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions and Future Work",
                "sec_num": "10."
            },
            {
                "text": "Finally, the overall goal of our research is to use phrase-level sentiment analysis in higher-level NLP tasks, such as opinion question answering and summarization.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusions and Future Work",
                "sec_num": "10."
            },
            {
                "text": "It is open to question whether reason should be listed as positive in a sentiment lexicon, because the more frequent senses of reason involve intention, not sentiment. However, any existing sentiment lexicon one would start with will have some noise and errors. The task in this article is to disambiguate instances of the entries in a given sentiment lexicon.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Available at http://nrrc.mitre.org/NRRC/publications.htm.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Some sentences contain additional subjective expressions that are not underlined as examples.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Both annotators are authors of this article. 5 The revised version of the MPQA corpus with the contextual polarity annotations is available at http://www.cs.pitt.edu/mpqa. 6 The third annotator received training until her reliability of performance on the task was comparable to that of the first two annotators who participated in the study.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "In the end, about 70% of the words from the General Inquirer positive word list and 80% of the words from the negative word list were included in the subjectivity lexicon.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "We decided on a different polarity for about 80 of the words in our lexicon that appeared on other positive and negative word lists.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Although the negation features give the best performance improvements of the three feature sets, these classifiers still do not perform as well as the respective all-feature polarity classifiers for each algorithm.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "To clarify, Section 8.2.3 only reported results for instances identified as polar in step one. Here, we report results for all clue instances, including the instances classified as neutral in step one.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "InGrefenstette et al. (2004), the units that are classified are fixed windows around named entities rather than sentences.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "Das and Chen (2001),Pang, Lee, and Vaithyanathan (2002), andDave, Lawrence, and Pennock (2003) also represent negation. In their experiments, words which follow a negation term are tagged with a negation marker and then treated as new words. Pang, Lee and Vaithyanathan report that representing negation in this way slightly helps their results, whereas Dave, Lawrence, and Pennock report a slightly detrimental effect.Whitelaw, Garg, and Argamon (2005) also represent negation terms and intensifiers. However, in their experiments, the effect of negation is not separately evaluated, and intensifiers are not found to be beneficial.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "We would like to thank the anonymous reviewers for their valuable comments and suggestions. This work was supported in part by an Andrew Mellow Predoctoral Fellowship, by the NSF under grant IIS-0208798, by the Advanced Research and Development Activity (ARDA), and by the European IST Programme through the AMIDA Integrated Project FP6-0033812.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Mining WordNet for fuzzy sentiment: Sentiment tag extraction from WordNet glosses",
                "authors": [
                    {
                        "first": "Alina",
                        "middle": [],
                        "last": "Andreevskaia",
                        "suffix": ""
                    },
                    {
                        "first": "Sabine",
                        "middle": [],
                        "last": "Bergler",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the 11th Meeting of the European Chapter of the Association for Computational Linguistics (EACL-2006)",
                "volume": "",
                "issue": "",
                "pages": "209--216",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Andreevskaia, Alina and Sabine Bergler. 2006. Mining WordNet for fuzzy sentiment: Sentiment tag extraction from WordNet glosses. In Proceedings of the 11th Meeting of the European Chapter of the Association for Computational Linguistics (EACL-2006), pages 209-216, Trento. Bai, Xue, Rema Padman, and Edoardo Airoldi. 2005. On learning parsimonious models for extracting consumer opinions. In Proceedings of the 38th Annual Hawaii International Conference on System Sciences (HICSS'05) -Track 3, page 75.2, Waikoloa, HI.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "The sentimental factor: Improving review classification via human-provided information",
                "authors": [
                    {
                        "first": "Ann",
                        "middle": [],
                        "last": "Banfield",
                        "suffix": ""
                    },
                    {
                        "first": "Kegan",
                        "middle": [],
                        "last": "Routledge",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Paul",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Boston",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Beineke",
                        "suffix": ""
                    },
                    {
                        "first": "Trevor",
                        "middle": [],
                        "last": "Philip",
                        "suffix": ""
                    },
                    {
                        "first": "Shivakumar",
                        "middle": [],
                        "last": "Hastie",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Vaithyanathan",
                        "suffix": ""
                    }
                ],
                "year": 1982,
                "venue": "Proceedings of the 42nd Annual Meeting of the Association for Computational Linguistics (ACL-04)",
                "volume": "",
                "issue": "",
                "pages": "263--270",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Banfield, Ann. 1982. Unspeakable Sentences. Routledge and Kegan Paul, Boston. Beineke, Philip, Trevor Hastie, and Shivakumar Vaithyanathan. 2004. The sentimental factor: Improving review classification via human-provided information. In Proceedings of the 42nd Annual Meeting of the Association for Computational Linguistics (ACL-04), pages 263-270, Barcelona.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Learning trees and rules with set-valued features",
                "authors": [
                    {
                        "first": "William",
                        "middle": [
                            "W"
                        ],
                        "last": "Cohen",
                        "suffix": ""
                    }
                ],
                "year": 1996,
                "venue": "Proceedings of the 13th National Conference on Artificial Intelligence",
                "volume": "",
                "issue": "",
                "pages": "709--717",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Cohen, William W. 1996. Learning trees and rules with set-valued features. In Proceedings of the 13th National Conference on Artificial Intelligence, pages 709-717, Portland, OR.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Combined optimization of feature selection and algorithm parameter interaction in machine learning of language",
                "authors": [
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Collins",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "Proceedings of the 35th Annual Meeting of the Association for Computational Linguistics (ACL-97)",
                "volume": "",
                "issue": "",
                "pages": "16--23",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Collins, Michael. 1997. Three generative, lexicalised models for statistical parsing. In Proceedings of the 35th Annual Meeting of the Association for Computational Linguistics (ACL-97), pages 16-23, Madrid. Daelemans, Walter, V\u00e9ronique Hoste, Fien De Meulder, and Bart Naudts. 2003a. Combined optimization of feature selection and algorithm parameter interaction in machine learning of language. In Proceedings of the 14th",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "European Conference on Machine Learning (ECML-2003)",
                "authors": [],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "84--95",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "European Conference on Machine Learning (ECML-2003), pages 84-95, Cavtat-Dubrovnik.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "TiMBL: Tilburg Memory Based Learner, version 5.0 Reference Guide",
                "authors": [
                    {
                        "first": "Walter",
                        "middle": [],
                        "last": "Daelemans",
                        "suffix": ""
                    },
                    {
                        "first": "Jakub",
                        "middle": [],
                        "last": "Zavrel",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Ko Van Der",
                        "suffix": ""
                    },
                    {
                        "first": "Antal",
                        "middle": [],
                        "last": "Sloot",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Van Den",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Bosch",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Daelemans, Walter, Jakub Zavrel, Ko van der Sloot, and Antal van den Bosch. 2003b. TiMBL: Tilburg Memory Based Learner, version 5.0 Reference Guide. ILK Technical Report 03-10, Induction of Linguistic Knowledge Research Group, Tilburg University. Available at http://ilk.uvt. nl/downloads/pub/papers/ilk0310.pdf.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Yahoo! for Amazon: Sentiment parsing from small talk on the Web",
                "authors": [
                    {
                        "first": "Sanjiv",
                        "middle": [],
                        "last": "Das",
                        "suffix": ""
                    },
                    {
                        "first": "Mike",
                        "middle": [
                            "Y"
                        ],
                        "last": "Ranjan",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Chen",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the August 2001 Meeting of the European Finance Association (EFA)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Das, Sanjiv Ranjan and Mike Y. Chen. 2001. Yahoo! for Amazon: Sentiment parsing from small talk on the Web. In Proceedings of the August 2001 Meeting of the European Finance Association (EFA), Barcelona, Spain. Available at http://ssrn.com/ abstract=276189.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Mining the peanut gallery: Opinion extraction and semantic classification of product reviews",
                "authors": [
                    {
                        "first": "Kushal",
                        "middle": [],
                        "last": "Dave",
                        "suffix": ""
                    },
                    {
                        "first": "Steve",
                        "middle": [],
                        "last": "Lawrence",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [
                            "M"
                        ],
                        "last": "Pennock",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the 12th International World Wide Web Conference (WWW2003)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dave, Kushal, Steve Lawrence, and David M. Pennock. 2003. Mining the peanut gallery: Opinion extraction and semantic classification of product reviews. In Proceedings of the 12th International World Wide Web Conference (WWW2003), Budapest. Available at http://www2003.org.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Determining the semantic orientation of terms through gloss analysis",
                "authors": [
                    {
                        "first": "Andrea",
                        "middle": [],
                        "last": "Esuli",
                        "suffix": ""
                    },
                    {
                        "first": "Fabrizio",
                        "middle": [],
                        "last": "Sebastiani",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of ACM SIGIR Conference on Information and Knowledge Management (CIKM-05)",
                "volume": "",
                "issue": "",
                "pages": "617--624",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Esuli, Andrea and Fabrizio Sebastiani. 2005. Determining the semantic orientation of terms through gloss analysis. In Proceedings of ACM SIGIR Conference on Information and Knowledge Management (CIKM-05), pages 617-624, Bremen.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Determining term subjectivity and term orientation for opinion mining",
                "authors": [
                    {
                        "first": "Andrea",
                        "middle": [],
                        "last": "Esuli",
                        "suffix": ""
                    },
                    {
                        "first": "Fabrizio",
                        "middle": [],
                        "last": "Sebastiani",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings the 11th Meeting of the European Chapter of the Association for Computational Linguistics (EACL-2006)",
                "volume": "",
                "issue": "",
                "pages": "193--200",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Esuli, Andrea and Fabrizio Sebastiani. 2006a. Determining term subjectivity and term orientation for opinion mining. In Proceedings the 11th Meeting of the European Chapter of the Association for Computational Linguistics (EACL-2006), pages 193-200, Trento.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "SentiWordNet: A publicly available lexical resource for opinion mining",
                "authors": [
                    {
                        "first": "Andrea",
                        "middle": [],
                        "last": "Esuli",
                        "suffix": ""
                    },
                    {
                        "first": "Fabrizio",
                        "middle": [],
                        "last": "Sebastiani",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of LREC-06, the 5th Conference on Language Resources and Evaluation",
                "volume": "",
                "issue": "",
                "pages": "417--422",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Esuli, Andrea and Fabrizio Sebastiani. 2006b. SentiWordNet: A publicly available lexical resource for opinion mining. In Proceedings of LREC-06, the 5th Conference on Language Resources and Evaluation, pages 417-422, Genoa.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Sentiment classification on customer feedback data: Noisy data, large feature vectors, and the role of linguistic analysis",
                "authors": [
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Gamon",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the 20th International Conference on Computational Linguistics (COLING-2004)",
                "volume": "",
                "issue": "",
                "pages": "611--617",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gamon, Michael. 2004. Sentiment classification on customer feedback data: Noisy data, large feature vectors, and the role of linguistic analysis. In Proceedings of the 20th International Conference on Computational Linguistics (COLING-2004), pages 611-617, Geneva.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Coupling niche browsers and affect analysis for an opinion mining application",
                "authors": [
                    {
                        "first": "Gregory",
                        "middle": [],
                        "last": "Grefenstette",
                        "suffix": ""
                    },
                    {
                        "first": "Yan",
                        "middle": [],
                        "last": "Qu",
                        "suffix": ""
                    },
                    {
                        "first": "James",
                        "middle": [
                            "G"
                        ],
                        "last": "Shanahan",
                        "suffix": ""
                    },
                    {
                        "first": "David",
                        "middle": [
                            "A"
                        ],
                        "last": "Evans",
                        "suffix": ""
                    },
                    {
                        "first": ";",
                        "middle": [],
                        "last": "",
                        "suffix": ""
                    },
                    {
                        "first": "Kathy",
                        "middle": [],
                        "last": "Mckeown",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "Proceedings of the 35th Annual Meeting of the Association for Computational Linguistics (ACL-97)",
                "volume": "",
                "issue": "",
                "pages": "174--181",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Grefenstette, Gregory, Yan Qu, James G. Shanahan, and David A. Evans. 2004. Coupling niche browsers and affect analysis for an opinion mining application. In Proceedings of the Conference Recherche d'Information Assistee par Ordinateur (RIAO-2004), pages 186-194, Avignon. Hatzivassiloglou, Vasileios and Kathy McKeown. 1997. Predicting the semantic orientation of adjectives. In Proceedings of the 35th Annual Meeting of the Association for Computational Linguistics (ACL-97), pages 174-181, Madrid.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "Optimization Issues in Machine Learning of Coreference Resolution",
                "authors": [
                    {
                        "first": "V\u00e9ronique",
                        "middle": [],
                        "last": "Hoste",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hoste, V\u00e9ronique. 2005. Optimization Issues in Machine Learning of Coreference Resolution. Ph.D. thesis, Language Technology Group, University of Antwerp.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Mining and summarizing customer reviews",
                "authors": [
                    {
                        "first": "Minqing",
                        "middle": [],
                        "last": "Hu",
                        "suffix": ""
                    },
                    {
                        "first": "Bing",
                        "middle": [],
                        "last": "Liu",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of ACM SIGKDD Conference on Knowledge Discovery and Data Mining",
                "volume": "",
                "issue": "",
                "pages": "168--177",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hu, Minqing and Bing Liu. 2004. Mining and summarizing customer reviews. In Proceedings of ACM SIGKDD Conference on Knowledge Discovery and Data Mining 2004 (KDD-2004), pages 168-177, Seattle, WA.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "On the foundations of relaxation labeling processes",
                "authors": [
                    {
                        "first": "Robert",
                        "middle": [
                            "A"
                        ],
                        "last": "Hummel",
                        "suffix": ""
                    },
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Steven",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Zucker",
                        "suffix": ""
                    }
                ],
                "year": 1983,
                "venue": "IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI)",
                "volume": "5",
                "issue": "3",
                "pages": "167--187",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Hummel, Robert A. and Steven W. Zucker. 1983. On the foundations of relaxation labeling processes. IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), 5(3):167-187.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Making large-scale SVM learning practical",
                "authors": [
                    {
                        "first": "Thorsten",
                        "middle": [],
                        "last": "Joachims",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Advances in Kernel Methods -Support Vector Learning",
                "volume": "",
                "issue": "",
                "pages": "169--184",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Joachims, Thorsten. 1999. Making large-scale SVM learning practical. In B. Scholkopf, C. Burgess, and A. Smola, editors, Advances in Kernel Methods -Support Vector Learning, pages 169-184. MIT Press, Cambridge, MA.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Automatic construction of polarity-tagged corpus from HTML documents",
                "authors": [
                    {
                        "first": "Nobuhiro",
                        "middle": [],
                        "last": "Kaji",
                        "suffix": ""
                    },
                    {
                        "first": "Masaru",
                        "middle": [],
                        "last": "Kitsuregawa",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions",
                "volume": "",
                "issue": "",
                "pages": "332--341",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kaji, Nobuhiro and Masaru Kitsuregawa. 2006. Automatic construction of polarity-tagged corpus from HTML documents. In Proceedings of the COLING/ACL 2006 Main Conference Poster Sessions, pages 452-459, Sydney. Kamps, Jaap and Maarten Marx. 2002. Words with attitude. In Proceedings of the 1st International Conference on Global WordNet, pages 332-341, Mysore.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Fully automatic lexicon expansion for domain-oriented sentiment analysis",
                "authors": [
                    {
                        "first": "Hiroshi",
                        "middle": [],
                        "last": "Kanayama",
                        "suffix": ""
                    },
                    {
                        "first": "Tetsuya",
                        "middle": [],
                        "last": "Nasukawa",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2006)",
                "volume": "",
                "issue": "",
                "pages": "355--363",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kanayama, Hiroshi and Tetsuya Nasukawa. 2006. Fully automatic lexicon expansion for domain-oriented sentiment analysis. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2006), pages 355-363, Sydney.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "Sentiment classification of movie reviews using contextual valence shifters",
                "authors": [
                    {
                        "first": "Alistair",
                        "middle": [],
                        "last": "Kennedy",
                        "suffix": ""
                    },
                    {
                        "first": "Diana",
                        "middle": [],
                        "last": "Inkpen",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Computational Intelligence",
                "volume": "22",
                "issue": "2",
                "pages": "110--125",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kennedy, Alistair and Diana Inkpen. 2006. Sentiment classification of movie reviews using contextual valence shifters. Computational Intelligence, 22(2):110-125.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Determining the sentiment of opinions",
                "authors": [
                    {
                        "first": "Soo-Min",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Eduard",
                        "middle": [],
                        "last": "Hovy",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the 20th International Conference on Computational Linguistics (COLING-2004)",
                "volume": "",
                "issue": "",
                "pages": "1267--1373",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Kim, Soo-Min and Eduard Hovy. 2004. Determining the sentiment of opinions. In Proceedings of the 20th International Conference on Computational Linguistics (COLING-2004), pages 1267-1373, Geneva.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "The importance of neutral examples for learning sentiment",
                "authors": [
                    {
                        "first": "Moshe",
                        "middle": [],
                        "last": "Koppel",
                        "suffix": ""
                    },
                    {
                        "first": "Jonathan",
                        "middle": [],
                        "last": "Schler",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Computational Intelligence",
                "volume": "22",
                "issue": "2",
                "pages": "100--109",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Koppel, Moshe and Jonathan Schler. 2006. The importance of neutral examples for learning sentiment. Computational Intelligence, 22(2):100-109.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "New Directions in Question Answering",
                "authors": [
                    {
                        "first": "Mark",
                        "middle": [
                            "T"
                        ],
                        "last": "Maybury",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "American Association for Artificial Intelligence",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Maybury, Mark T., editor. 2004. New Directions in Question Answering. American Association for Artificial Intelligence, Menlo Park, CA.",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Mining product reputations on the Web",
                "authors": [
                    {
                        "first": "Satoshi",
                        "middle": [],
                        "last": "Morinaga",
                        "suffix": ""
                    },
                    {
                        "first": "Kenji",
                        "middle": [],
                        "last": "Yamanishi",
                        "suffix": ""
                    },
                    {
                        "first": "Kenji",
                        "middle": [],
                        "last": "Tateishi",
                        "suffix": ""
                    },
                    {
                        "first": "Toshikazu",
                        "middle": [],
                        "last": "Fukushima",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2002)",
                "volume": "",
                "issue": "",
                "pages": "341--349",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Morinaga, Satoshi, Kenji Yamanishi, Kenji Tateishi, and Toshikazu Fukushima. 2002. Mining product reputations on the Web. In Proceedings of the 8th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD-2002), pages 341-349, Edmonton.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "Sentiment analysis using support vector machines with diverse information sources",
                "authors": [
                    {
                        "first": "Tony",
                        "middle": [],
                        "last": "Mullen",
                        "suffix": ""
                    },
                    {
                        "first": "Nigel",
                        "middle": [],
                        "last": "Collier",
                        "suffix": ""
                    },
                    {
                        "first": ";",
                        "middle": [],
                        "last": "Barcelona",
                        "suffix": ""
                    },
                    {
                        "first": "Tetsuya",
                        "middle": [],
                        "last": "Nasukawa",
                        "suffix": ""
                    },
                    {
                        "first": "Jeonghee",
                        "middle": [],
                        "last": "Yi",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2004)",
                "volume": "",
                "issue": "",
                "pages": "79--86",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mullen, Tony and Nigel Collier. 2004. Sentiment analysis using support vector machines with diverse information sources. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2004), pages 412-418, Barcelona. Nasukawa, Tetsuya and Jeonghee Yi. 2003. Sentiment analysis: Capturing favorability using natural language processing. In Proceedings of the 2nd International Conference on Knowledge Capture (K-CAP 2003), pages 70-77, Sanibel Island, FL. Pang, Bo, Lillian Lee, and Shivakumar Vaithyanathan. 2002. Thumbs up? Sentiment classification using machine learning techniques. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2002), pages 79-86, Philadelphia, PA.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "Extracting product features and opinions from reviews",
                "authors": [
                    {
                        "first": "Livia",
                        "middle": [],
                        "last": "Polanyi",
                        "suffix": ""
                    },
                    {
                        "first": "Annie",
                        "middle": [],
                        "last": "Zaenen",
                        "suffix": ""
                    },
                    {
                        "first": ";",
                        "middle": [],
                        "last": "Popescu",
                        "suffix": ""
                    },
                    {
                        "first": "Ana-Maria",
                        "middle": [],
                        "last": "",
                        "suffix": ""
                    },
                    {
                        "first": "Oren",
                        "middle": [],
                        "last": "Etzioni",
                        "suffix": ""
                    }
                ],
                "year": 1985,
                "venue": "Proceedings of the Human Language Technologies Conference/Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP-2005)",
                "volume": "",
                "issue": "",
                "pages": "339--346",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Polanyi, Livia and Annie Zaenen. 2004. Contextual valence shifters. In Working Notes of the AAAI Spring Symposium on Exploring Attitude and Affect in Text: Theories and Applications, pages 106-111, The AAAI Press, Menlo Park, CA. Popescu, Ana-Maria and Oren Etzioni. 2005. Extracting product features and opinions from reviews. In Proceedings of the Human Language Technologies Conference/Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP-2005), pages 339-346, Vancouver. Quirk, Randolph, Sidney Greenbaum, Geoffry Leech, and Jan Svartvik. 1985. A Comprehensive Grammar of the English Language. Longman, New York.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Learning extraction patterns for subjective expressions",
                "authors": [
                    {
                        "first": "Ellen",
                        "middle": [],
                        "last": "Riloff",
                        "suffix": ""
                    },
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2003)",
                "volume": "",
                "issue": "",
                "pages": "105--112",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Riloff, Ellen and Janyce Wiebe. 2003. Learning extraction patterns for subjective expressions. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2003), pages 105-112, Sapporo.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "BoosTexter: A boosting-based system for text categorization",
                "authors": [
                    {
                        "first": "Robert",
                        "middle": [
                            "E"
                        ],
                        "last": "Schapire",
                        "suffix": ""
                    },
                    {
                        "first": "Yoram",
                        "middle": [],
                        "last": "Singer",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "Machine Learning",
                "volume": "39",
                "issue": "",
                "pages": "135--168",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Schapire, Robert E. and Yoram Singer. 2000. BoosTexter: A boosting-based system for text categorization. Machine Learning, 39(2/3):135-168.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "Smokey: Automatic recognition of hostile messages",
                "authors": [
                    {
                        "first": "Ellen",
                        "middle": [],
                        "last": "Spertus",
                        "suffix": ""
                    }
                ],
                "year": 1997,
                "venue": "Proceedings of the 8th Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-97)",
                "volume": "",
                "issue": "",
                "pages": "1058--1065",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Spertus, Ellen. 1997. Smokey: Automatic recognition of hostile messages. In Proceedings of the 8th Annual Conference on Innovative Applications of Artificial Intelligence (IAAI-97), pages 1058-1065, Providence, RI.",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "The General Inquirer: A Computer Approach to Content Analysis",
                "authors": [
                    {
                        "first": "Philip",
                        "middle": [
                            "J"
                        ],
                        "last": "Stone",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Dexter",
                        "suffix": ""
                    },
                    {
                        "first": "Marshall",
                        "middle": [
                            "S"
                        ],
                        "last": "Dunphy",
                        "suffix": ""
                    },
                    {
                        "first": "Daniel",
                        "middle": [
                            "M"
                        ],
                        "last": "Smith",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Ogilvie",
                        "suffix": ""
                    }
                ],
                "year": 1966,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Stone, Philip J., Dexter C. Dunphy, Marshall S. Smith, and Daniel M. Ogilvie. 1966. The General Inquirer: A Computer Approach to Content Analysis. MIT Press, Cambridge, MA.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Multi-perspective question answering using the OpQA corpus",
                "authors": [
                    {
                        "first": "Veselin",
                        "middle": [],
                        "last": "Stoyanov",
                        "suffix": ""
                    },
                    {
                        "first": "Claire",
                        "middle": [],
                        "last": "Cardie",
                        "suffix": ""
                    },
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the Human Language Technologies Conference/ Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP-2005)",
                "volume": "",
                "issue": "",
                "pages": "923--930",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Stoyanov, Veselin, Claire Cardie, and Janyce Wiebe. 2005. Multi-perspective question answering using the OpQA corpus. In Proceedings of the Human Language Technologies Conference/ Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP-2005), pages 923-930, Vancouver.",
                "links": null
            },
            "BIBREF31": {
                "ref_id": "b31",
                "title": "Application of semi-supervised learning to evaluative expression classification",
                "authors": [
                    {
                        "first": "Yasuhiro",
                        "middle": [],
                        "last": "Suzuki",
                        "suffix": ""
                    },
                    {
                        "first": "Hiroya",
                        "middle": [],
                        "last": "Takamura",
                        "suffix": ""
                    },
                    {
                        "first": "Manabu",
                        "middle": [],
                        "last": "Okumura",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the 7th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2006)",
                "volume": "",
                "issue": "",
                "pages": "502--513",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Suzuki, Yasuhiro, Hiroya Takamura, and Manabu Okumura. 2006. Application of semi-supervised learning to evaluative expression classification. In Proceedings of the 7th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2006), pages 502-513, Mexico City.",
                "links": null
            },
            "BIBREF32": {
                "ref_id": "b32",
                "title": "Extracting emotional polarity of words using spin model",
                "authors": [
                    {
                        "first": "Hiroya",
                        "middle": [],
                        "last": "Takamura",
                        "suffix": ""
                    },
                    {
                        "first": "Takashi",
                        "middle": [],
                        "last": "Inui",
                        "suffix": ""
                    },
                    {
                        "first": "Manabu",
                        "middle": [],
                        "last": "Okumura",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL-05)",
                "volume": "",
                "issue": "",
                "pages": "133--140",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Takamura, Hiroya, Takashi Inui, and Manabu Okumura. 2005. Extracting emotional polarity of words using spin model. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL-05), pages 133-140, Ann Arbor, MI.",
                "links": null
            },
            "BIBREF33": {
                "ref_id": "b33",
                "title": "An operational system for detecting and tracking opinions in online discussions",
                "authors": [
                    {
                        "first": "Richard",
                        "middle": [],
                        "last": "Tong",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Working Notes of the SIGIR Workshop on Operational Text Classification",
                "volume": "",
                "issue": "",
                "pages": "1--6",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Tong, Richard. 2001. An operational system for detecting and tracking opinions in online discussions. In Working Notes of the SIGIR Workshop on Operational Text Classification, pages 1-6, New Orleans, LA.",
                "links": null
            },
            "BIBREF34": {
                "ref_id": "b34",
                "title": "Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews",
                "authors": [
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Turney",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (ACL-02)",
                "volume": "",
                "issue": "",
                "pages": "417--424",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Turney, Peter. 2002. Thumbs up or thumbs down? Semantic orientation applied to unsupervised classification of reviews. In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics (ACL-02), pages 417-424, Philadelphia, PA.",
                "links": null
            },
            "BIBREF35": {
                "ref_id": "b35",
                "title": "Measuring praise and criticism: Inference of semantic orientation from association",
                "authors": [
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Turney",
                        "suffix": ""
                    },
                    {
                        "first": "Michael",
                        "middle": [
                            "L"
                        ],
                        "last": "Littman",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "ACM Transactions on Information Systems (TOIS)",
                "volume": "21",
                "issue": "4",
                "pages": "315--346",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Turney, Peter and Michael L. Littman. 2003. Measuring praise and criticism: Inference of semantic orientation from association. ACM Transactions on Information Systems (TOIS), 21(4):315-346.",
                "links": null
            },
            "BIBREF36": {
                "ref_id": "b36",
                "title": "Using appraisal groups for sentiment analysis",
                "authors": [
                    {
                        "first": "Casey",
                        "middle": [],
                        "last": "Whitelaw",
                        "suffix": ""
                    },
                    {
                        "first": "Navendu",
                        "middle": [],
                        "last": "Garg",
                        "suffix": ""
                    },
                    {
                        "first": "Shlomo",
                        "middle": [],
                        "last": "Argamon",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the 14th ACM International Conference on Information and Knowledge Wilson, Wiebe, and Hoffmann Recognizing Contextual Polarity Management (CIKM-2005)",
                "volume": "",
                "issue": "",
                "pages": "625--631",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Whitelaw, Casey, Navendu Garg, and Shlomo Argamon. 2005. Using appraisal groups for sentiment analysis. In Proceedings of the 14th ACM International Conference on Information and Knowledge Wilson, Wiebe, and Hoffmann Recognizing Contextual Polarity Management (CIKM-2005), pages 625-631, Bremen.",
                "links": null
            },
            "BIBREF37": {
                "ref_id": "b37",
                "title": "Tracking point of view in narrative",
                "authors": [
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    }
                ],
                "year": 1994,
                "venue": "Computational Linguistics",
                "volume": "20",
                "issue": "2",
                "pages": "233--287",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wiebe, Janyce. 1994. Tracking point of view in narrative. Computational Linguistics, 20(2):233-287.",
                "links": null
            },
            "BIBREF38": {
                "ref_id": "b38",
                "title": "Development and use of a gold standard data set for subjectivity classifications",
                "authors": [
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    },
                    {
                        "first": "Rebecca",
                        "middle": [],
                        "last": "Bruce",
                        "suffix": ""
                    },
                    {
                        "first": "Thomas O'",
                        "middle": [],
                        "last": "Hara",
                        "suffix": ""
                    }
                ],
                "year": 1999,
                "venue": "Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics (ACL-99)",
                "volume": "",
                "issue": "",
                "pages": "246--253",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wiebe, Janyce, Rebecca Bruce, and Thomas O'Hara. 1999. Development and use of a gold standard data set for subjectivity classifications. In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics (ACL-99), pages 246-253, College Park, MD.",
                "links": null
            },
            "BIBREF39": {
                "ref_id": "b39",
                "title": "Word sense and subjectivity",
                "authors": [
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    },
                    {
                        "first": "Rada",
                        "middle": [],
                        "last": "Mihalcea",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "1065--1072",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wiebe, Janyce and Rada Mihalcea. 2006. Word sense and subjectivity. In Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, pages 1065-1072, Sydney.",
                "links": null
            },
            "BIBREF40": {
                "ref_id": "b40",
                "title": "Creating subjective and objective sentence classifiers from unannotated texts",
                "authors": [
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    },
                    {
                        "first": "Ellen",
                        "middle": [],
                        "last": "Riloff",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the 6th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2005)",
                "volume": "",
                "issue": "",
                "pages": "486--497",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wiebe, Janyce and Ellen Riloff. 2005. Creating subjective and objective sentence classifiers from unannotated texts. In Proceedings of the 6th International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-2005), pages 486-497, Mexico City.",
                "links": null
            },
            "BIBREF41": {
                "ref_id": "b41",
                "title": "Annotating expressions of opinions and emotions in language. Language Resources and Evaluation (formerly",
                "authors": [
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    },
                    {
                        "first": "Theresa",
                        "middle": [],
                        "last": "Wilson",
                        "suffix": ""
                    },
                    {
                        "first": "Claire",
                        "middle": [],
                        "last": "Cardie",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Computers and the Humanities)",
                "volume": "39",
                "issue": "2/3",
                "pages": "164--210",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wiebe, Janyce, Theresa Wilson, and Claire Cardie. 2005. Annotating expressions of opinions and emotions in language. Language Resources and Evaluation (formerly Computers and the Humanities), 39(2/3):164-210.",
                "links": null
            },
            "BIBREF42": {
                "ref_id": "b42",
                "title": "Recognizing contextual polarity in phrase-level sentiment analysis",
                "authors": [
                    {
                        "first": "Theresa",
                        "middle": [],
                        "last": "Wilson",
                        "suffix": ""
                    },
                    {
                        "first": "Janyce",
                        "middle": [],
                        "last": "Wiebe",
                        "suffix": ""
                    },
                    {
                        "first": "Paul",
                        "middle": [],
                        "last": "Hoffmann",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the Human Language Technologies Conference/ Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP-2005)",
                "volume": "",
                "issue": "",
                "pages": "347--354",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wilson, Theresa, Janyce Wiebe, and Paul Hoffmann. 2005. Recognizing contextual polarity in phrase-level sentiment analysis. In Proceedings of the Human Language Technologies Conference/ Conference on Empirical Methods in Natural Language Processing (HLT/EMNLP-2005), pages 347-354, Vancouver.",
                "links": null
            },
            "BIBREF43": {
                "ref_id": "b43",
                "title": "Converting dependency structures to phrase structures",
                "authors": [
                    {
                        "first": "Fei",
                        "middle": [],
                        "last": "Xia",
                        "suffix": ""
                    },
                    {
                        "first": "Martha",
                        "middle": [],
                        "last": "Palmer",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Proceedings of the Human Language Technology Conference (HLT-2001)",
                "volume": "",
                "issue": "",
                "pages": "1--5",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Xia, Fei and Martha Palmer. 2001. Converting dependency structures to phrase structures. In Proceedings of the Human Language Technology Conference (HLT-2001), pages 1-5, San Diego, CA.",
                "links": null
            },
            "BIBREF44": {
                "ref_id": "b44",
                "title": "Sentiment analyzer: Extracting sentiments about a given topic using natural language processing techniques",
                "authors": [
                    {
                        "first": "Jeonghee",
                        "middle": [],
                        "last": "Yi",
                        "suffix": ""
                    },
                    {
                        "first": "Tetsuya",
                        "middle": [],
                        "last": "Nasukawa",
                        "suffix": ""
                    },
                    {
                        "first": "Razvan",
                        "middle": [],
                        "last": "Bunescu",
                        "suffix": ""
                    },
                    {
                        "first": "Wayne",
                        "middle": [],
                        "last": "Niblack",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the 3rd IEEE International Conference on Data Mining (ICDM'03)",
                "volume": "",
                "issue": "",
                "pages": "427--434",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yi, Jeonghee, Tetsuya Nasukawa, Razvan Bunescu, and Wayne Niblack. 2003. Sentiment analyzer: Extracting sentiments about a given topic using natural language processing techniques. In Proceedings of the 3rd IEEE International Conference on Data Mining (ICDM'03), pages 427-434, Melbourne, FL.",
                "links": null
            },
            "BIBREF45": {
                "ref_id": "b45",
                "title": "Towards answering opinion questions: Separating facts from opinions and identifying the polarity of opinion sentences",
                "authors": [
                    {
                        "first": "Hong",
                        "middle": [],
                        "last": "Yu",
                        "suffix": ""
                    },
                    {
                        "first": "Vasileios",
                        "middle": [],
                        "last": "Hatzivassiloglou",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2003)",
                "volume": "",
                "issue": "",
                "pages": "129--136",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yu, Hong and Vasileios Hatzivassiloglou. 2003. Towards answering opinion questions: Separating facts from opinions and identifying the polarity of opinion sentences. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP-2003), pages 129-136, Sapporo.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "type_str": "figure",
                "num": null,
                "text": "Thousands of coup supporters celebrated (positive) overnight, waving flags, blowing whistles . . . (8) The criteria set by Rice are the following: the three countries in question are repressive (negative) and grave human rights violators (negative) . . . (9) Besides, politicians refer to good and evil (both) only for purposes of intimidation and exaggeration.",
                "uris": null
            },
            "FIGREF1": {
                "type_str": "figure",
                "num": null,
                "text": "Figure 2 The dependency tree for the sentence The human rights report poses a substantial challenge to the U.S. interpretation of good and evil. Prior polarity and reliability class are marked in parentheses for words that match clues from the lexicon.",
                "uris": null
            },
            "FIGREF2": {
                "type_str": "figure",
                "num": null,
                "text": "for a given metric as compared to the word+priorpol baseline (word baseline for Ripper) are indicated by + or -, respectively; ++ or --indicates the change is significant at the p < 0.1 level; +++ or ---indicates significance at the p < 0.05 level; nc indicates no change.",
                "uris": null
            },
            "TABREF0": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>victory of justice and freedom</td><td>such a disadvantageous situation</td></tr><tr><td>grown tremendously</td><td>must</td></tr><tr><td>such animosity</td><td>not true at all</td></tr><tr><td>throttling the voice</td><td>imperative for harmonious society</td></tr><tr><td>disdain and wrath</td><td>glorious</td></tr><tr><td>so exciting</td><td>disastrous consequences</td></tr><tr><td>could not have wished for a better situation</td><td>believes</td></tr><tr><td>freak show</td><td>the embodiment of two-sided justice</td></tr><tr><td>if you're not with us, you're against us</td><td>appalling</td></tr><tr><td>vehemently denied</td><td>very definitely</td></tr><tr><td>everything good and nice</td><td>once and for all</td></tr><tr><td>under no circumstances</td><td>shameful mum</td></tr><tr><td>most fraudulent, terrorist and extremist</td><td>enthusiastically asked</td></tr><tr><td>number one democracy</td><td>hate</td></tr><tr><td>seems to think</td><td>gross misstatement</td></tr><tr><td colspan=\"2\">indulging in blood-shed and their lunaticism surprised, to put it mildly</td></tr><tr><td>take justice to pre-historic times</td><td>unconditionally and without delay</td></tr><tr><td colspan=\"2\">so conservative that it makes Pat Buchanan look vegetarian</td></tr><tr><td colspan=\"2\">those digging graves for others, get engraved themselves</td></tr><tr><td colspan=\"2\">lost the reputation of commitment to principles of human justice</td></tr><tr><td colspan=\"2\">ultimately the demon they have reared will eat up their own vitals</td></tr></table>",
                "text": "Sample of subjective expressions from the MPQA corpus."
            },
            "TABREF1": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td/><td>Positive sentiments</td><td>Negative sentiments</td></tr><tr><td>Emotion</td><td>I'm happy</td><td>I'm sad</td></tr><tr><td colspan=\"2\">Evaluation Great idea!</td><td>Bad idea!</td></tr><tr><td>Stance</td><td colspan=\"2\">She supports the bill She's against the bill</td></tr></table>",
                "text": "Examples of positive and negative sentiments."
            },
            "TABREF2": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td/><td colspan=\"5\">Neutral Positive Negative Both Total</td></tr><tr><td>Neutral</td><td>123</td><td>14</td><td>24</td><td>0</td><td>161</td></tr><tr><td>Positive</td><td>16</td><td>73</td><td>5</td><td>2</td><td>9 6</td></tr><tr><td>Negative</td><td>14</td><td>2</td><td>167</td><td>1</td><td>184</td></tr><tr><td>Both</td><td>0</td><td>3</td><td>0</td><td>3</td><td>6</td></tr><tr><td>Total</td><td>153</td><td>92</td><td>196</td><td>6</td><td>447</td></tr><tr><td>Table 4</td><td/><td/><td/><td/><td/></tr><tr><td colspan=\"6\">Contingency table for contextual polarity agreement, borderline cases removed.</td></tr><tr><td/><td colspan=\"5\">Neutral Positive Negative Both Total</td></tr><tr><td>Neutral</td><td>113</td><td>7</td><td>8</td><td>0</td><td>128</td></tr><tr><td>Positive</td><td>9</td><td>59</td><td>3</td><td>0</td><td>7 1</td></tr><tr><td>Negative</td><td>5</td><td>2</td><td>156</td><td>1</td><td>164</td></tr><tr><td>Both</td><td>0</td><td>2</td><td>0</td><td>2</td><td>4</td></tr><tr><td>Total</td><td>127</td><td>70</td><td>167</td><td>3</td><td>367</td></tr></table>",
                "text": "Contingency table for contextual polarity agreement."
            },
            "TABREF3": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td colspan=\"4\">Neutral Positive Negative Both</td><td>Total</td></tr><tr><td>9,057</td><td>3,311</td><td>7,294</td><td>299</td><td>19,961</td></tr><tr><td>45.4%</td><td>16.6%</td><td>36.5%</td><td colspan=\"2\">1.5% 100%</td></tr></table>",
                "text": "Distribution of contextual polarity tags."
            },
            "TABREF5": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Word Features</td></tr><tr><td>word token</td></tr><tr><td>word part of speech</td></tr><tr><td>previous word part of speech</td></tr><tr><td>next word part of speech</td></tr><tr><td>prior polarity: positive, negative, both, neutral</td></tr><tr><td>reliability class: strongsubj or weaksubj</td></tr><tr><td>General Modification Features</td></tr><tr><td>preceded by adjective: binary</td></tr><tr><td>preceded by adverb (other than not): binary</td></tr><tr><td>preceded by intensifier: binary</td></tr><tr><td>self intensifier: binary</td></tr><tr><td>modifies strongsubj: binary</td></tr><tr><td>modifies weaksubj: binary</td></tr><tr><td>modified by strongsubj: binary</td></tr><tr><td>modified by weaksubj: binary</td></tr><tr><td>Polarity Modification Features</td></tr><tr><td>modifies polarity: Structure Features</td></tr><tr><td>in subject: binary</td></tr><tr><td>in copular: binary</td></tr><tr><td>in passive: binary</td></tr></table>",
                "text": "Features for neutral-polar classification. positive, negative, neutral, both, notmod modified by polarity: positive, negative, neutral, both, notmod conjunction polarity: positive, negative, neutral, both, notmod"
            },
            "TABREF6": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Topic</td><td>Description</td></tr><tr><td>argentina</td><td>Economic collapse in Argentina</td></tr><tr><td>axisofevil</td><td>U.S. President's State of the Union Address</td></tr><tr><td colspan=\"2\">guantanamo Detention of prisoners in Guantanamo Bay</td></tr><tr><td colspan=\"2\">humanrights U.S. State Department Human Rights Report</td></tr><tr><td>kyoto</td><td>Kyoto Protocol ratification</td></tr><tr><td>settlements</td><td>Israeli settlements in Gaza and the West Bank</td></tr><tr><td>space</td><td>Space missions of various countries</td></tr><tr><td>taiwan</td><td>Relationship between Taiwan and China</td></tr><tr><td>venezuela</td><td>Presidential coup in Venezuela</td></tr><tr><td>zimbabwe</td><td>Presidential election in Zimbabwe</td></tr></table>",
                "text": "Topics in the MPQA corpus."
            },
            "TABREF7": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Word Features</td></tr><tr><td>word token</td></tr><tr><td>word prior polarity: positive, negative, both, neutral</td></tr><tr><td>Negation Features</td></tr><tr><td>negated: binary</td></tr><tr><td>negated subject: binary</td></tr><tr><td>Polarity Modification Features</td></tr><tr><td>modifies polarity: positive, negative, neutral, both, notmod</td></tr><tr><td>modified by polarity: positive, negative, neutral, both, notmod</td></tr><tr><td>conj polarity: positive, negative, neutral, both, notmod</td></tr><tr><td>Polarity Shifters</td></tr><tr><td>general polarity shifter: binary</td></tr><tr><td>negative polarity shifter: binary</td></tr><tr><td>positive polarity shifter: binary</td></tr></table>",
                "text": "Features for polarity classification."
            },
            "TABREF8": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Algorithm</td><td>Settings</td></tr><tr><td colspan=\"2\">BoosTexter 2,000 rounds of boosting</td></tr><tr><td>TiMBL</td><td>k=25, MVDM distance metric</td></tr><tr><td>Ripper</td><td>-!n, -S 0.5</td></tr><tr><td>SVM</td><td>linear kernel</td></tr><tr><td>8.1.1</td><td/></tr></table>",
                "text": "Algorithm settings for neutral-polar classification. Classification Results. The results for the first set of experiments are given in"
            },
            "TABREF10": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Experiment</td><td>Features</td></tr><tr><td>PARTS-OF-SPEECH</td><td>parts of speech for clue instance, previous word, and next word</td></tr><tr><td>RELIABILITY-CLASS</td><td>reliability class of clue instance</td></tr><tr><td>PRECEDED-POS</td><td>preceded by adjective, preceded by adverb</td></tr><tr><td>INTENSIFY</td><td>preceded by intensifier, self intensifier</td></tr><tr><td>RELCLASS-MOD</td><td>modifies strongsubj/weaksubj, modified by strongsubj/weaksubj</td></tr><tr><td>POLARITY-MOD</td><td>polarity-modification features</td></tr><tr><td>STRUCTURE</td><td>structure features</td></tr><tr><td>CURSENT-COUNTS</td><td>strongsubj/weaksubj clue instances in sentence</td></tr><tr><td>PNSENT-COUNTS</td><td>strongsubj/weaksubj clue instances in previous/next sentence</td></tr><tr><td>CURSENT-OTHER</td><td>adjectives/adverbs/cardinal number/pronoun/modal in sentence</td></tr></table>",
                "text": "Neutral-polar feature sets for evaluation."
            },
            "TABREF11": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td/><td/><td colspan=\"2\">Polar Neut</td><td/><td/><td colspan=\"2\">Polar Neut</td></tr><tr><td>BoosTexter</td><td>Acc</td><td>F</td><td>F</td><td>Ripper</td><td>Acc</td><td>F</td><td>F</td></tr><tr><td>PARTS-OF-SPEECH</td><td>+</td><td>-</td><td>+</td><td>PARTS-OF-SPEECH</td><td>+++</td><td>+++</td><td>---</td></tr><tr><td>RELIABILITY-CLASS</td><td>+</td><td>-</td><td>+</td><td>RELIABILITY-CLASS</td><td>+++</td><td>+++</td><td>+</td></tr><tr><td>PRECEDED-POS</td><td>nc</td><td>-</td><td>nc</td><td>PRECEDED-POS</td><td/><td/><td/></tr></table>",
                "text": "Results for neutral-polar feature ablation experiments."
            },
            "TABREF13": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td>Experiment</td><td/><td/><td/><td/><td>Features</td><td/><td/></tr><tr><td>NEGATION</td><td colspan=\"4\">negated, negated subject</td><td/><td/><td/></tr><tr><td>POLARITY-MOD</td><td colspan=\"7\">modifies polarity, modified by polarity, conjunction polarity</td></tr><tr><td>SHIFTERS</td><td colspan=\"6\">general, negative, positive polarity shifters</td><td/></tr><tr><td>Table 17</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"6\">Results for polarity feature ablation experiments.</td><td/><td/></tr><tr><td/><td/><td/><td>Positive</td><td/><td colspan=\"2\">Negative</td><td/></tr><tr><td/><td colspan=\"2\">Acc Rec</td><td colspan=\"2\">Prec F</td><td>Rec</td><td colspan=\"2\">Prec F</td></tr><tr><td>BoosTexter</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>NEGATION</td><td colspan=\"2\">+++ ++</td><td colspan=\"4\">+++ +++ +++ +</td><td>+++</td></tr><tr><td>POLARITY-MOD</td><td>++</td><td colspan=\"2\">+++ +</td><td colspan=\"2\">+++ +</td><td>++</td><td>+</td></tr><tr><td>SHIFTERS</td><td>+</td><td>+</td><td>+</td><td>+</td><td>+</td><td>+</td><td>+</td></tr><tr><td>TiMBL</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>NEGATION</td><td colspan=\"7\">+++ +++ +++ +++ +++ +++ +++</td></tr><tr><td>POLARITY-MOD</td><td>+</td><td>+</td><td>+</td><td>+</td><td>-</td><td>+</td><td>+</td></tr><tr><td>SHIFTERS</td><td>+</td><td>+</td><td>+</td><td>+</td><td>-</td><td>+</td><td>+</td></tr><tr><td>Ripper</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>NEGATION</td><td colspan=\"2\">+++ --</td><td colspan=\"4\">+++ +++ +++ -</td><td>+++</td></tr><tr><td>POLARITY-MOD</td><td>+</td><td colspan=\"2\">+++ ++</td><td colspan=\"2\">+++ +</td><td>+</td><td>+</td></tr><tr><td>SHIFTERS</td><td>+</td><td>-</td><td>+</td><td>+</td><td>+</td><td>-</td><td>+</td></tr><tr><td>SVM</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>NEGATION</td><td colspan=\"2\">+++ -</td><td colspan=\"4\">+++ +++ +++ +</td><td>+++</td></tr><tr><td>POLARITY-MOD</td><td>+</td><td>-</td><td colspan=\"2\">+++ +</td><td>+</td><td>-</td><td>+</td></tr><tr><td>SHIFTERS</td><td>+</td><td>-</td><td>+</td><td>+</td><td>+</td><td>+</td><td>+</td></tr></table>",
                "text": "Polarity feature sets for evaluation."
            },
            "TABREF14": {
                "type_str": "table",
                "html": null,
                "num": null,
                "content": "<table><tr><td/><td colspan=\"4\">Acc Pos F Neg F Both F</td></tr><tr><td>BoosTexter</td><td/><td/><td/><td/></tr><tr><td colspan=\"2\">excluding word token 82.5</td><td>74.9</td><td>88.0</td><td>17.4</td></tr><tr><td>all polarity features</td><td>83.2</td><td>75.5</td><td>88.7</td><td>19.4</td></tr><tr><td>TiMBL</td><td/><td/><td/><td/></tr><tr><td colspan=\"2\">excluding word token 83.2</td><td>75.9</td><td>88.4</td><td>17.3</td></tr><tr><td>all polarity features</td><td>82.2</td><td>74.3</td><td>88.0</td><td>23.9</td></tr><tr><td>Ripper</td><td/><td/><td/><td/></tr><tr><td colspan=\"2\">excluding word token 82.9</td><td>75.4</td><td>88.3</td><td>17.4</td></tr><tr><td>all polarity features</td><td>83.2</td><td>75.6</td><td>88.5</td><td>17.4</td></tr><tr><td>SVM</td><td/><td/><td/><td/></tr><tr><td colspan=\"2\">excluding word token 81.5</td><td>72.9</td><td>87.3</td><td>16.8</td></tr><tr><td>all polarity features</td><td>81.6</td><td>72.9</td><td>87.3</td><td>16.9</td></tr></table>",
                "text": "Results for polarity classification without and with the word token feature."
            }
        }
    }
}