File size: 104,463 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
{
    "paper_id": "D09-1013",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T16:39:58.090391Z"
    },
    "title": "A Rich Feature Vector for Protein-Protein Interaction Extraction from Multiple Corpora",
    "authors": [
        {
            "first": "Makoto",
            "middle": [],
            "last": "Miwa",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "the University of Tokyo",
                "location": {
                    "addrLine": "Japan Hongo 7-3-1, Bunkyo-ku",
                    "settlement": "Tokyo",
                    "country": "Japan"
                }
            },
            "email": "mmiwa@is.s.u-tokyo.ac.jp"
        },
        {
            "first": "Rune",
            "middle": [],
            "last": "Saetre",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "the University of Tokyo",
                "location": {
                    "addrLine": "Japan Hongo 7-3-1, Bunkyo-ku",
                    "settlement": "Tokyo",
                    "country": "Japan"
                }
            },
            "email": "rune.saetre@is.s.u-tokyo.ac.jp"
        },
        {
            "first": "Yusuke",
            "middle": [],
            "last": "Miyao",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "the University of Tokyo",
                "location": {
                    "addrLine": "Japan Hongo 7-3-1, Bunkyo-ku",
                    "settlement": "Tokyo",
                    "country": "Japan"
                }
            },
            "email": "yusuke@is.s.u-tokyo.ac.jp"
        },
        {
            "first": "Jun",
            "middle": [
                "'"
            ],
            "last": "Ichi Tsujii",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "the University of Tokyo",
                "location": {
                    "addrLine": "Japan Hongo 7-3-1, Bunkyo-ku",
                    "settlement": "Tokyo",
                    "country": "Japan"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Because of the importance of proteinprotein interaction (PPI) extraction from text, many corpora have been proposed with slightly differing definitions of proteins and PPI. Since no single corpus is large enough to saturate a machine learning system, it is necessary to learn from multiple different corpora. In this paper, we propose a solution to this challenge. We designed a rich feature vector, and we applied a support vector machine modified for corpus weighting (SVM-CW) to complete the task of multiple corpora PPI extraction. The rich feature vector, made from multiple useful kernels, is used to express the important information for PPI extraction, and the system with our feature vector was shown to be both faster and more accurate than the original kernelbased system, even when using just a single corpus. SVM-CW learns from one corpus, while using other corpora for support. SVM-CW is simple, but it is more effective than other methods that have been successfully applied to other NLP tasks earlier. With the feature vector and SVM-CW, our system achieved the best performance among all state-of-the-art PPI extraction systems reported so far.",
    "pdf_parse": {
        "paper_id": "D09-1013",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Because of the importance of proteinprotein interaction (PPI) extraction from text, many corpora have been proposed with slightly differing definitions of proteins and PPI. Since no single corpus is large enough to saturate a machine learning system, it is necessary to learn from multiple different corpora. In this paper, we propose a solution to this challenge. We designed a rich feature vector, and we applied a support vector machine modified for corpus weighting (SVM-CW) to complete the task of multiple corpora PPI extraction. The rich feature vector, made from multiple useful kernels, is used to express the important information for PPI extraction, and the system with our feature vector was shown to be both faster and more accurate than the original kernelbased system, even when using just a single corpus. SVM-CW learns from one corpus, while using other corpora for support. SVM-CW is simple, but it is more effective than other methods that have been successfully applied to other NLP tasks earlier. With the feature vector and SVM-CW, our system achieved the best performance among all state-of-the-art PPI extraction systems reported so far.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "The performance of an information extraction program is highly dependent on various factors, including text types (abstracts, complete articles, reports, etc.), exact definitions of the information to be extracted, shared sub-topics of the text collections from which information is to be extracted.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Even if two corpora are annotated in terms of the same type of information by two groups, the performance of a program trained by one corpus is unlikely to be reproduced in the other corpus. On the other hand, from a practical point of view, it is worth while to effectively use multiple existing annotated corpora together, because it is very costly to make new annotations.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "One problem with several different corpora is protein-protein interaction (PPI) extraction from text. While PPIs play a critical role in understanding the working of cells in diverse biological contexts, the manual construction of PPI databases such as BIND, DIP, HPRD, IntAct, and MINT (Mathivanan et al., 2006 ) is known to be very time-consuming and labor-intensive. The automatic extraction of PPI from published papers has therefore been a major research topic in Natural Language Processing for Biology (BioNLP). Among several PPI extraction task settings, the most common is sentence-based, pair-wise PPI extraction. At least four annotated corpora have been provided for this setting: AIMed (Bunescu et al., 2005) , HPRD50 (Fundel et al., 2006) , IEPA (Ding et al., 2002) , and LLL (N\u00e9dellec, 2005) . Each of these corpora have been used as the standard corpus for training and testing PPI programs. Moreover, several corpora are annotated for more types of events than just for PPI. Such examples include BioInfer (Pyysalo et al., 2007) , and GENIA (Kim et al., 2008a) , and they can be reorganized into PPI corpora. Even though all of these corpora were made for PPI extraction, they were constructed based on different definitions of proteins and PPI, which reflect different biological research interests .",
                "cite_spans": [
                    {
                        "start": 264,
                        "end": 311,
                        "text": "HPRD, IntAct, and MINT (Mathivanan et al., 2006",
                        "ref_id": null
                    },
                    {
                        "start": 699,
                        "end": 721,
                        "text": "(Bunescu et al., 2005)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 731,
                        "end": 752,
                        "text": "(Fundel et al., 2006)",
                        "ref_id": "BIBREF14"
                    },
                    {
                        "start": 760,
                        "end": 779,
                        "text": "(Ding et al., 2002)",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 790,
                        "end": 806,
                        "text": "(N\u00e9dellec, 2005)",
                        "ref_id": "BIBREF23"
                    },
                    {
                        "start": 1023,
                        "end": 1045,
                        "text": "(Pyysalo et al., 2007)",
                        "ref_id": "BIBREF25"
                    },
                    {
                        "start": 1058,
                        "end": 1077,
                        "text": "(Kim et al., 2008a)",
                        "ref_id": "BIBREF17"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Research on PPI extraction so far has revealed that the performance on each of the corpora could benefit from additional examples . Learning from multiple annotated corpora could lead to better PPI extraction performance. Various research paradigms such as inductive transfer learning (ITL) and domain adaptation (DA) have mainly focused on how to effectively use corpora annotated by other groups, by reducing the incompatibilities (Pan and Yang, 2008) .",
                "cite_spans": [
                    {
                        "start": 442,
                        "end": 453,
                        "text": "Yang, 2008)",
                        "ref_id": "BIBREF24"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this paper, we propose the extraction of PPIs from multiple different corpora. We design a rich feature vector, and as an ITL method, we apply a support vector machine (SVM) modified for corpus weighting (SVM-CW) (Schweikert et al., 2008) , in order to evaluate the use of multiple corpora for the PPI extraction task. Our rich feature vector is made from multiple useful kernels, each of which is based on multiple parser inputs, proposed by Miwa et al. (2008) . The system with our feature vector was better than or at least comparable to the state-of-the-art PPI extraction systems on every corpus. The system is a good starting point to use the multiple corpora. Using one of the corpora as the target corpus, SVM-CW weights the remaining corpora (we call them the source corpora) with \"goodness\" for training on the target corpus. While SVM-CW is simple, we show that SVM-CW can improve the performance of the system more effectively and more efficiently than other methods proven to be successful in other NLP tasks earlier. As a result, SVM-CW with our feature vector is comprised of a PPI system with five different models, of which each model is superior to the best model in the original PPI extraction task, which used only the single corpus.",
                "cite_spans": [
                    {
                        "start": 216,
                        "end": 241,
                        "text": "(Schweikert et al., 2008)",
                        "ref_id": "BIBREF29"
                    },
                    {
                        "start": 446,
                        "end": 464,
                        "text": "Miwa et al. (2008)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "While sentence-based, pair-wise PPI extraction was initially tackled by using simple methods based on co-occurrences, lately, more sophisticated machine learning systems augmented by NLP techniques have been applied (Bunescu et al., 2005) . The task has been tackled as a classification problem. To pull out useful information from NLP tools including taggers and parsers, several kernels have been applied to calculate the similarity between PPI pairs. Miwa et al. (2008) recently proposed the use of multiple kernels using multiple parsers. This outperformed other systems on the AIMed, which is the most frequently used corpus for the PPI extraction task, by a wide margin.",
                "cite_spans": [
                    {
                        "start": 216,
                        "end": 238,
                        "text": "(Bunescu et al., 2005)",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 454,
                        "end": 472,
                        "text": "Miwa et al. (2008)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Works",
                "sec_num": "2"
            },
            {
                "text": "To improve the performance using external C  l  a  s  s  i  f  i  c  a  t  i  o  n  R  e  s  u  l  t   T  r  a  i  n  i  n  g  D  a  t  a   F  e  a  t  u  r  e  v  e  c  t  o  r   R  a  w  T  e  x  t  s   P  a  r  s  e  r  s  C  l  a  s  s  i  f  i  e  r   T  e  s  t  D  a  t  a   R  a  w  T  e  x  t  s   M  o  d  e  l   P  a  i  r  I  n  f  o  r  m  a  t  i  o  n   P  a  i  r  I  n  f  o  r  m  a  t  i  o  n   L  a  b e l Figure 1 : Overview of our PPI extraction system training data, many ITL and DA methods have been proposed. Most of ITL methods assume that the feature space is same, and that the labels may be different in only some examples, while most of DA methods assume that the labels are the same, and that the feature space is different. Among the methods, we use adaptive SVM (aSVM) , singular value decomposition (SVD) based alternating structure optimization (SVD-ASO) (Ando et al., 2005) , and transfer AdaBoost (TrAdaBoost) (Dai et al., 2007) to compare with SVM-CW. We do not use semi-supervised learning (SSL) methods, because it would be considerably costly to generate enough clean unlabeled data needed for SSL (Erkan et al., 2007) . aSVM is seen as a promising DA method among several modifications of SVM including SVM-CW. aSVM tries to find a model that is close to the one made from other classification problems. SVD-ASO is one of the most successful SSL, DA, or multi-task learning methods in NLP. The method tries to find an additional useful feature space by solving auxiliary problems that are close to the target problem. With well-designed auxiliary problems, the method has been applied to text classification, text chunking, and word sense disambiguation (Ando, 2006) . The method was reported to perform better than or comparable to the best state-of-the-art systems in all of these tasks. TrAd-aBoost was proposed as an ITL method. In training, the method reduces the effect of incompatible examples by decreasing their weights, and thereby tries to use useful examples from source corpora. The method has been applied to text classification, and the reported performance was better than SVM and transductive SVM (Dai et al., 2007) .",
                "cite_spans": [
                    {
                        "start": 891,
                        "end": 910,
                        "text": "(Ando et al., 2005)",
                        "ref_id": "BIBREF2"
                    },
                    {
                        "start": 948,
                        "end": 966,
                        "text": "(Dai et al., 2007)",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 1140,
                        "end": 1160,
                        "text": "(Erkan et al., 2007)",
                        "ref_id": "BIBREF12"
                    },
                    {
                        "start": 1697,
                        "end": 1709,
                        "text": "(Ando, 2006)",
                        "ref_id": "BIBREF3"
                    },
                    {
                        "start": 2157,
                        "end": 2175,
                        "text": "(Dai et al., 2007)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 42,
                        "end": 422,
                        "text": "C  l  a  s  s  i  f  i  c  a  t  i  o  n  R  e  s  u  l  t   T  r  a  i  n  i  n  g  D  a  t  a   F  e  a  t  u  r  e  v  e  c  t  o  r   R  a  w  T  e  x  t  s   P  a  r  s  e  r  s  C  l  a  s  s  i  f  i  e  r   T  e  s  t  D  a  t  a   R  a  w  T  e  x  t  s   M  o  d  e  l   P  a  i  r  I  n  f  o  r  m  a  t  i  o  n   P  a  i  r  I  n  f  o  r  m  a  t  i  o  n   L  a  b",
                        "ref_id": null
                    },
                    {
                        "start": 427,
                        "end": 435,
                        "text": "Figure 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Related Works",
                "sec_num": "2"
            },
            {
                "text": "The target task of our system is a sentence-based, pair-wise PPI extraction. It is formulated as a classification problem that judges whether a given pair XPG p1 protein interacts with multiple subunits of TFIIH prot and with CSB p2 protein. Figure 1 shows the overview of the proposed PPI extraction system. As a classifier using a single corpus, we use the 2-norm soft-margin linear SVM (L2-SVM) classifier, with the dual coordinate decent (DCD) method, by . In this section, we explain the two main features: the feature vector, and the corpus weighting method for multiple corpora.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 242,
                        "end": 250,
                        "text": "Figure 1",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "PPI Extraction System",
                "sec_num": "3"
            },
            {
                "text": "We propose a feature vector with three types of features, corresponding to the three different kernels, which were each combined with the two parsers: the Enju 2.3.0, and KSDEP beta 1 ; this feature vector is used because the kernels with these parsers were shown to be effective for PPI extraction by Miwa et al. (2008) , and because it is important to start from a good performance single corpus system. Both parsers were retrained using the GENIA Treebank corpus provided by Kim et al. (2003) . By using our linear feature vector, we can perform calculations faster by using fast linear classifiers like L2-SVM, and we also obtain a more accurate extraction, than by using the original kernel method. output is grouped according to the feature-type and parser, and each group of features is separately normalized by the L2-norm 1 . Finally, all values are put into a single feature vector, and the whole feature vector is then also normalized by the L2norm. The features are constructed by using predicate argument structures (PAS) from Enju, and by using the dependency trees from KSDEP.",
                "cite_spans": [
                    {
                        "start": 302,
                        "end": 320,
                        "text": "Miwa et al. (2008)",
                        "ref_id": "BIBREF21"
                    },
                    {
                        "start": 478,
                        "end": 495,
                        "text": "Kim et al. (2003)",
                        "ref_id": "BIBREF16"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Vector",
                "sec_num": "3.1"
            },
            {
                "text": "N M O D S B J r N M O D E N T I T Y 1 p r o t e i n i n t e r a c t E N T I T Y 2 p r o t e i n p r o t e i n E N T I T Y 1 p r o t e i n i n t e r a c t s w i t h m u l t i p l e a n d w i t h E N T I T Y 2 p r o t e i n . N M O D S B J C O O D C O O R D N M O D P M O D N M O D S B J r N M O D p r o t e i n i n t e r a c t p r o t e i n S B J r C O O D r P M O D V - w a l k s E - w a l k s \u30fb \u30fb \u30fb \u30fb \u30fb \u30fb \u30fb \u30fb \u30fb",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Feature Vector",
                "sec_num": "3.1"
            },
            {
                "text": "The BOW feature includes the lemma form of a word, its relative position to the target pair of proteins (Before, Middle, After), and its frequency in the target sentence. BOW features form the BOW kernel in the original kernel method. BOW features for the pair in Figure 2 are shown in Figure 4 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 264,
                        "end": 272,
                        "text": "Figure 2",
                        "ref_id": "FIGREF0"
                    },
                    {
                        "start": 286,
                        "end": 294,
                        "text": "Figure 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Bag-of-Words (BOW) Features",
                "sec_num": "3.1.1"
            },
            {
                "text": "SP features include vertex walks (v-walks), edge walks (e-walks), and their subsets (Kim et al., 2008b) on the target pair in a parse structure, and represent the connection between the pair. The features are the subsets of the tree kernels on the shortest path (Saetre et al., 2007) . an e-walk includes a lemma and its two links. The links indicates the predicate argument relations for PAS, and the dependencies for dependency trees.",
                "cite_spans": [
                    {
                        "start": 84,
                        "end": 103,
                        "text": "(Kim et al., 2008b)",
                        "ref_id": "BIBREF18"
                    },
                    {
                        "start": 262,
                        "end": 283,
                        "text": "(Saetre et al., 2007)",
                        "ref_id": "BIBREF28"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Shortest Path (SP) Features",
                "sec_num": "3.1.2"
            },
            {
                "text": "Graph features are made from the all-paths graph kernel proposed by . The kernel represents the target pair using graph matrices based on two subgraphs, and the graph features are all the non-zero elements in the graph matrices.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Graph Features",
                "sec_num": "3.1.3"
            },
            {
                "text": "The two subgraphs are a parse structure subgraph (PSS) and a linear order subgraph (LOS). Figure 6 describes the subgraphs of the sentence parsed by KSDEP in Figure 2 . PSS represents the parse structure of a sentence. PSS has word vertices or link vertices. A word vertex contains its lemma and its part-of-speech (POS), while a link vertex contains its link. Additionally, both types of vertices contain their positions relative to the shortest path. The \"IP\"s in the vertices on the shortest path represent the positions, and the vertices are differentiated from the other vertices like \"P\", \"CC\", and \"and:CC\" in Figure 6 . LOS represents the word sequence in the sentence. LOS has word vertices, each of which contains its lemma, its relative position to the target pair, and its POS.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 90,
                        "end": 98,
                        "text": "Figure 6",
                        "ref_id": null
                    },
                    {
                        "start": 158,
                        "end": 166,
                        "text": "Figure 2",
                        "ref_id": "FIGREF0"
                    },
                    {
                        "start": 617,
                        "end": 625,
                        "text": "Figure 6",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Graph Features",
                "sec_num": "3.1.3"
            },
            {
                "text": "Each subgraph is represented by a graph matrix G as follows:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Graph Features",
                "sec_num": "3.1.3"
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "G = L T \u221e n=1 A n L,",
                        "eq_num": "(1)"
                    }
                ],
                "section": "Graph Features",
                "sec_num": "3.1.3"
            },
            {
                "text": "where L is a N \u00d7L label matrix, A is an N \u00d7N edge matrix, N represents the number of vertices, and L represents the number of labels. The label of a vertex includes all information described above (e.g. \"ENTITY1:NN:IP\" in Figure 6 ). If two vertices have exactly same information, the labels will be same. G can be calculated efficiently by using the Neumann Series . The label matrix represents the correspondence between labels and vertices. L ij is 1 if the i-th vertex corresponds to the j-th label, and 0 otherwise. The edge matrix represents the connection between the pairs of vertices. A ij is a weight w ij (0.9 or 0.3 in Figure 6 ) if the i-th vertex is connected to the j-th vertex, and 0 otherwise. By this calculation, G ij represent the sum of the weights of all paths between the i-th label and the j-th label. Figure 7: Learning curves on two large corpora. The x-axis is related to the percentage of the examples in a corpus. The curves are obtained by a 10-fold CV with a random split. Table 1 shows the sizes of the PPI corpora that we used. Their widely-ranged differences including the sizes were manually analyzed by . While AIMed, HPRD50, IEPA, and LLL were all annotated as PPI corpora, BioInfer in its original form contains much more fine-grained information than does just the PPI. BioInfer was transformed into a PPI corpus by a program, so making it the largest of the five. Among them, AIMed alone was created by annotating whole abstracts, while the other corpora were made by annotating single sentences selected from abstracts. Figure 7 shows the learning curves on two large corpora: AIMed and BioInfer. The curves are obtained by performing a 10-fold cross validation (CV) on each corpus, with random splits, using our system. The curves show that the performances can benefit from the additional examples. To get a better PPI extraction system for a chosen target, we need to draw useful shared information from external source corpora. We refer to examples in the source corpora as \"source examples\", and examples in a target corpus as \"target examples\". Among the corpora, we assume that the labels in some examples are incompatible, and that their distributions are also different, but that the feature space is shared.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 222,
                        "end": 230,
                        "text": "Figure 6",
                        "ref_id": null
                    },
                    {
                        "start": 631,
                        "end": 639,
                        "text": "Figure 6",
                        "ref_id": null
                    },
                    {
                        "start": 1004,
                        "end": 1011,
                        "text": "Table 1",
                        "ref_id": "TABREF1"
                    },
                    {
                        "start": 1561,
                        "end": 1569,
                        "text": "Figure 7",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Graph Features",
                "sec_num": "3.1.3"
            },
            {
                "text": "In order to draw useful information from the source corpora to get a better model for the target Figure 6 : Parse structure subgraph and linear order subgraph to extract graph features of the pair in Figure 2 . The parse structure subgraph is from the parse tree by KSDEP.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 97,
                        "end": 105,
                        "text": "Figure 6",
                        "ref_id": null
                    },
                    {
                        "start": 200,
                        "end": 208,
                        "text": "Figure 2",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "E N T I T Y 1 N N I P p r o t e i n N N I P i n t e r a c t V B Z I P w i t h I N I P m u l t i p l e J J s u b u n i t N N S o f I N P R O T N N a n d C C w i t h I N I P E N T I T Y 2 N N I P p r o t e i n N N I P . . N M O D I P S B J I P C O O D I P P M O D N M O D N M O D P M O D C C C O O R D I P N M O D I P P M O D I P P E N T I T Y 1 N N p r o t e i n N N M i n t e r a c t V B Z M w i t h I N M m u l t i p l e J J M s u b u n i t N N S M o f I N M P R O T N N M a n d C C M w i t h I N M E N T I T Y 2 N",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "corpus, we use SVM-CW, which has been used as a DA method. Given a set of instance-label pairs (x i , y i ), i = 1, . . ., ls + lt, x i \u2208R n , and y i \u2208{\u22121, +1}, we solve the following problem:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "EQUATION",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [
                    {
                        "start": 0,
                        "end": 8,
                        "text": "EQUATION",
                        "ref_id": "EQREF",
                        "raw_str": "min w 1 2 w T w + C s ls i=1 i + C t ls+lt i=ls+1 i ,",
                        "eq_num": "(2)"
                    }
                ],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "where w is a weight vector, is a loss function, and ls and lt are the numbers of source and target examples respectively. C s \u2265 0 and C t \u2265 0 are penalty parameters. We use a squared hinge loss",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "i = max(0, 1 \u2212 y i w T x i ) 2 .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "Here, the source corpora are treated as one corpus. The problem, excluding the second term, is equal to L2-SVM. The problem can be solved using the DCD method.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "As an ITL method, SVM-CW weights each corpus, and tries to benefit from the source corpora, by adjusting the effect of their compatibility and incompatibility. For the adjustment, these penalty parameters should be set properly. Since we are unaware of the widely ranged differences among the corpora, we empirically estimated them by performing 10-fold CV on the training data.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Corpus Weighting for Mixing Corpora",
                "sec_num": "3.2"
            },
            {
                "text": "We used five corpora for evaluation: AIMed, BioInfer, HPRD50, IEPA, and LLL. For the comparison with other methods, we report the Fscore (%), and the area under the receiver operating characteristic (ROC) curve (AUC) (%) using (abstract-wise) a 10-fold CV and a oneanswer-per-occurrence criterion. These measures are commonly used for the PPI extraction tasks. The F-score is a harmonic mean of Precision and Recall. The ROC curve is a plot of a true positive rate (TPR) vs a false positive rate (FPR) for different thresholds. We tuned the regularization parameters of all classifiers by performing a 10fold CV on the training data using a random split. The other parameters were fixed, and we report the highest of the macro-averaged F-scores as our final F-score. For 10-fold CV, we split the corpora as recommended by .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation Settings",
                "sec_num": "4.1"
            },
            {
                "text": "In this section, we evaluate our system on a single corpus, in order to evaluate our feature vector and to justify the use of the following modules: normalization methods and classification methods.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "PPI Extraction on a Single Corpus",
                "sec_num": "4.2"
            },
            {
                "text": "First, we compare our preprocessing method with other preprocessing methods to confirm how our preprocessing method improves the performance. Our method produced 64.2% in F-score using L2-SVM on AIMed. Scaling all features individually to have a maximal absolute value of 1, produced only 44.2% in the F-score, while normalizing the feature vector by L2-norm produced 61.5% in the F-score. Both methods were inferior to our method, because the values of features in the same group should be treated together, and Table 2 : Classification performance on AIMed using five different linear classifiers. The F-score (F) and Area Under the ROC curve (AUC) are shown. L2 is L2-SVM, L1 is L1-SVM, LR is logistic regression, AP is averaged perceptron, and CW is confidence weighted linear classification.",
                "cite_spans": [
                    {
                        "start": 612,
                        "end": 615,
                        "text": "(F)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 513,
                        "end": 520,
                        "text": "Table 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "PPI Extraction on a Single Corpus",
                "sec_num": "4.2"
            },
            {
                "text": "group with different values can produce better results, as will be explored in our future work.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "PPI Extraction on a Single Corpus",
                "sec_num": "4.2"
            },
            {
                "text": "Next, using our feature vector, we applied five different linear classifiers to extract PPI from AIMed: L2-SVM, 1-norm soft-margin SVM (L1-SVM), logistic regression (LR) (Fan et al., 2008) , averaged perceptron (AP) (Collins, 2002) , and confidence weighted linear classification (CW) (Dredze et al., 2008) . Table 2 indicates the performance of these classifiers on AIMed. We employed better settings for the task than did the original methods for AP and CW. We used a Widrow-Hoff learning rule (Bishop, 1995) for AP, and we performed one iteration for CW. L2-SVM is as good as, if not better, than other classifiers (Fscore and AUC). In the least, L2-SVM is as fast as these classifiers. AP and CW are worse than the other three methods, because they require a large number of examples, and are un-suitable for the current task. This result indicates that all linear classifiers, with the exception of AP and CW, perform almost equally, when using our feature vector.",
                "cite_spans": [
                    {
                        "start": 170,
                        "end": 188,
                        "text": "(Fan et al., 2008)",
                        "ref_id": "BIBREF13"
                    },
                    {
                        "start": 216,
                        "end": 231,
                        "text": "(Collins, 2002)",
                        "ref_id": "BIBREF8"
                    },
                    {
                        "start": 285,
                        "end": 306,
                        "text": "(Dredze et al., 2008)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 496,
                        "end": 510,
                        "text": "(Bishop, 1995)",
                        "ref_id": "BIBREF4"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 309,
                        "end": 316,
                        "text": "Table 2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "PPI Extraction on a Single Corpus",
                "sec_num": "4.2"
            },
            {
                "text": "Finally, we implemented the kernel method by Miwa et al. (2008) . For a 10-fold CV on AIMed, the running time was 9,507 seconds, and the performance was 61.5% F-score and 87.1% AUC. Our system used 4,702 seconds, and the performance was 64.2% F-score and 89.1% AUC. This result displayed that our system, with L2-SVM, and our new feature vector, is better, and faster, than the kernel-based system.",
                "cite_spans": [
                    {
                        "start": 45,
                        "end": 63,
                        "text": "Miwa et al. (2008)",
                        "ref_id": "BIBREF21"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "PPI Extraction on a Single Corpus",
                "sec_num": "4.2"
            },
            {
                "text": "In this section, we first apply each model from a source corpus to a target corpus, to show how different the corpora are. We then evaluate SVM-CW by comparing it with three other methods (see Section 2) with limited features, and apply it to every corpus. A  I  M  e  d  B  i  o  I  n  f  e  r  H  P  R  D  5  0  I  E  P  A  L  L  L   F   T  a  r  g  e  t  c  o  r  p  u  s   A  I  M  e  d  B  i  o  I  n  f  e  r  H  P  R  D  5 Figure 8 : F-score on a target corpus using a model on a source corpus. For the comparison, we show the 10-fold CV result on each target corpus and co-occurrences. The regularization parameter was fixed to 1.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 257,
                        "end": 429,
                        "text": "A  I  M  e  d  B  i  o  I  n  f  e  r  H  P  R  D  5  0  I  E  P  A  L  L  L   F   T  a  r  g  e  t  c  o  r  p  u  s   A  I  M  e  d  B  i  o  I  n  f  e  r  H  P  R  D  5",
                        "ref_id": null
                    },
                    {
                        "start": 430,
                        "end": 438,
                        "text": "Figure 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Corpus Weighting",
                "sec_num": "4.3"
            },
            {
                "text": "0 I E P A L L L c o - o c c M o d e l",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Evaluation of Corpus Weighting",
                "sec_num": "4.3"
            },
            {
                "text": "First, we apply the model from a source corpus to a target corpus. Figure 8 shows how the model from a source corpus performs on the target corpus. Interestingly, the model from IEPA performs better on LLL than the model from LLL itself. All the results showed that using different corpora (except IEPA) is worse than just using the same corpora. However, the cross-corpora scores are still better than the co-occurrences base-line, which indicates that the corpora share some information, even though they are not fully compatible.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 67,
                        "end": 75,
                        "text": "Figure 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Corpus Weighting",
                "sec_num": "4.3"
            },
            {
                "text": "Next, we compare SVM-CW with three other methods: aSVM, SVD-ASO, and TrAdaBoost. For this comparison, we used our feature vector without including the graph features, because SVD-ASO and TrAdaBoost require large computational resources. We applied SVD-ASO and TrAdaBoost in the following way. As for SVD-ASO, we made 400 auxiliary problems from the labels of each corpus by splitting features randomly, and extracted 50 additional features each for 4 feature groups. In total, we made new 200 additional features from 2,000 auxiliary problems. As recommended by Ando et al. 2005, we removed negative weights, performed SVD to each feature group, and iterated ASO once. Since Ad-aBoost easily overfitted with our rich feature vector, we applied soft margins (Ratsch et al., 2001) to TrAdaBoost. The update parameter for source examples was calculated using the update parameter on the training data in AdaBoost and the original parameter in TrAdaBoost. This ensures that the parameter would be the same as the original parameter, when the C value in the soft margin approaches infinity. Table 4 : F-score and AUC by SVM-CW. Rows correspond to a target corpus, and columns a source corpus. A:AIMed, B:BioInfer, H:HPRD50, I:IEPA, and L:LLL corpora. \"all\" signifies that all source corpora are used as one source corpus, ignoring the differences among the corpora. For the comparison, we show the 10-fold CV result on each target corpus.",
                "cite_spans": [
                    {
                        "start": 757,
                        "end": 778,
                        "text": "(Ratsch et al., 2001)",
                        "ref_id": "BIBREF27"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1086,
                        "end": 1093,
                        "text": "Table 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Corpus Weighting",
                "sec_num": "4.3"
            },
            {
                "text": "In Table 3 , we demonstrate the results of the comparison. SVM-CW improved the classification performance at least as much as all the other methods. The improvement is mainly attributed to the aggressive use of source examples while learning the model. Some source examples can be used as training data, as indicated in Figure 8 . SVM-CW does not set the restriction between C s and C t in Equation 2, so it can use source examples aggressively while learning the model. Since aSVM transfers a model, and SVD-ASO transfers an additional feature space, aSVM and SVD-ASO do not use the source examples while learning the model. In addition to the difference in the data usage, the settings of aSVM and SVD-ASO do not match the current task. As for aSVM, the DA assumption (that the labels are the same) does not match the task. In SVD-ASO, the numbers of both source examples and auxiliary problems are much smaller than those reported by Ando et al. (2005) . TrAdaBoost uses the source examples while learning the model, but never increases the weight of the examples, and it attempts to reduce their effects.",
                "cite_spans": [
                    {
                        "start": 937,
                        "end": 955,
                        "text": "Ando et al. (2005)",
                        "ref_id": "BIBREF2"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 3,
                        "end": 10,
                        "text": "Table 3",
                        "ref_id": "TABREF5"
                    },
                    {
                        "start": 320,
                        "end": 328,
                        "text": "Figure 8",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Corpus Weighting",
                "sec_num": "4.3"
            },
            {
                "text": "Finally, we apply SVM-CW to all corpora using all features. Table 4 summarizes the F-score and AUC by SVM-CW with all features. SVM-CW is especially effective for small corpora, showing that SVM-CW can adapt source corpora to a small annotated target corpus. The improvement on AIMed is small compared to the improvement on BioInfer, even though these corpora are similar in size. One of the reasons for this is that whole abstracts are annotated in AIMed, therefore making the examples biased. The difference between L2-SVM and SVM-CW + IEPA on AIMed is small, but statistically, it is significant (McNemar test (McNemar, 1947) , P = 0.0081). In the cases of HPRD50 + IEPA, LLL + IEPA, and two folds in BioInfer + IEPA, C s is larger than C t in Equation (2). This is worth noting, because the source corpus is more weighted than the target corpus, and the prediction performance on the target corpus is improved. Most methods put more trust in the target corpus than in the source corpus, and our results show that this setting is not always effective for mixing corpora. The results also indicate that IEPA contains more useful information for extracting PPI than other corpora, and that using source examples aggressively is important for these combinations. We compared the results of L2-SVM and SVM-CW + IEPA on AIMed, and found that 38 pairs were described as \"interaction\" or \"binding\" in the sentences among 61 . A:AIMed, B:BioInfer, H:HPRD50, I:IEPA, and L:LLL corpora. The results with the highest F-score from Table 4 are reported as the results for SVM-CW. newly found pairs. This analysis is evidence that IEPA contains instances to help find such interactions, and that SVM-CW helps to collect gold pairs that lack enough supporting instances in a single corpus, by adding instances from other corpora. SVM-CW missed coreferential relations that were also missed by L2-SVM. This can be attributed to the fact that the coreferential information is not stored in our current feature vector; so we need an even more expressive feature space. This is left as future work. SVM-CW is effective on most corpus combinations, and all the models from single corpora can be improved by adding other source corpora. This result is impressive, because the baselines by L2-SVM on just single corpora are already better than or at least comparable to other state-of-the-art PPI extraction systems, and also because the variety of the differences among different corpora is quite wide depending on various factors including annotation policies of the corpora . The results suggest that SVM-CW is useful as an ITL method.",
                "cite_spans": [
                    {
                        "start": 599,
                        "end": 628,
                        "text": "(McNemar test (McNemar, 1947)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 60,
                        "end": 67,
                        "text": "Table 4",
                        "ref_id": null
                    },
                    {
                        "start": 1522,
                        "end": 1529,
                        "text": "Table 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Evaluation of Corpus Weighting",
                "sec_num": "4.3"
            },
            {
                "text": "We compare our system with other previously published PPI extraction systems. Tables 5 and 6 summarize the comparison. Table 5 summarizes the comparison of several PPI extraction systems evaluated on the AIMed corpus. As indicated, the performance of the heavy kernel method is lower than our fast rich feature-vector method. Our system is, to the extent of our knowledge, the best performing PPI extraction system evaluated on the AIMed corpus, both in terms of AUC and F-scores. first reported results using all five corpora. We cannot directly com-pare our result with the F-score results, because they tuned the threshold, but our system still outperforms the system by on every corpus in AUC values. The results also indicate that our system outperforms other systems on all PPI corpora, and that both the rich feature vector and the corpus weighting are effective for the PPI extraction task.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 119,
                        "end": 126,
                        "text": "Table 5",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Comparison with Other PPI Systems",
                "sec_num": "4.4"
            },
            {
                "text": "In this paper, we proposed a PPI extraction system with a rich feature vector, using a corpus weighting method (SVM-CW) for combining the multiple PPI corpora. The feature vector extracts as much information as possible from the main training corpus, and SVM-CW incorporate other external source corpora in order to improve the performance of the classifier on the main target corpus. To the extent of our knowledge, this is the first application of ITL and DA methods to PPI extraction. As a result, the system, with SVM-CW and the feature vector, outperformed all other PPI extraction systems on all of the corpora. The PPI corpora share some information, and it is shown to be effective to add other source corpora when working with a specific target corpus.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "5"
            },
            {
                "text": "The main contributions of this paper are: 1) conducting experiments in extracting PPI using multiple corpora, 2) suggesting a rich feature vector using several previously proposed features and normalization methods, 3) the combination of SVM with corpus weighting and the new feature vector improved results on this task compared with prior work.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "5"
            },
            {
                "text": "There are many differences among the corpora that we used, and some of the differences are still unresolved. For further improvement, it would be necessary to investigate what is shared and what is different among the corpora. The SVM-CW method, and the PPI extraction system, can be applied generally to other classification tasks, and to other binary relation extraction tasks, without the need for modification. There are several other tasks in which many different corpora, which at first glance seem compatible, exist. By applying SVM-CW to such corpora, we will analyze which differences can be resolved by SVM-CW, and what differences require a manual resolution.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "5"
            },
            {
                "text": "For the PPI extraction system, we found many false negatives that need to be resolved. For further improvement, we need to analyze the cause",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "5"
            },
            {
                "text": "The vector normalized by the L2-norm is also called a unit vector.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "This work was partially supported by Grant-in-Aid for Specially Promoted Research (MEXT, Japan), Genome Network Project (MEXT, Japan), and Scientific Research (C) (General) (MEXT, Japan).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "The numbers of positive and all examples, precision (P), recall (R), F-score (F), and AUC are shown. The result with the highest F-score from Table 4 is reported as the result for SVM-CW. The scores in the parentheses of Miwa et al. (2008) indicate the result using the same 10-fold splits as our result",
                "authors": [],
                "year": null,
                "venue": "Comparison with previous PPI extraction results on the AIMed corpus",
                "volume": "5",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Table 5: Comparison with previous PPI extraction results on the AIMed corpus. The numbers of positive and all examples, precision (P), recall (R), F-score (F), and AUC are shown. The result with the highest F-score from Table 4 is reported as the result for SVM-CW. The scores in the parentheses of Miwa et al. (2008) indicate the result using the same 10-fold splits as our result, as indicated in Section 4.2.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "All-paths graph kernel for protein-protein interaction extraction with evaluation of cross corpus learning",
                "authors": [
                    {
                        "first": "Antti",
                        "middle": [],
                        "last": "Airola",
                        "suffix": ""
                    },
                    {
                        "first": "Sampo",
                        "middle": [],
                        "last": "Pyysalo",
                        "suffix": ""
                    },
                    {
                        "first": "Jari",
                        "middle": [],
                        "last": "Bj\u00f6rne",
                        "suffix": ""
                    },
                    {
                        "first": "Tapio",
                        "middle": [],
                        "last": "Pahikkala",
                        "suffix": ""
                    },
                    {
                        "first": "Filip",
                        "middle": [],
                        "last": "Ginter",
                        "suffix": ""
                    },
                    {
                        "first": "Tapio",
                        "middle": [],
                        "last": "Salakoski",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "BMC Bioinformatics",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Antti Airola, Sampo Pyysalo, Jari Bj\u00f6rne, Tapio Pahikkala, Filip Ginter, and Tapio Salakoski. 2008. All-paths graph kernel for protein-protein interac- tion extraction with evaluation of cross corpus learn- ing. BMC Bioinformatics.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "A framework for learning predictive structures from multiple tasks and unlabeled data",
                "authors": [
                    {
                        "first": "Tong",
                        "middle": [],
                        "last": "Rie Kubota Ando",
                        "suffix": ""
                    },
                    {
                        "first": "Peter",
                        "middle": [],
                        "last": "Zhang",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Bartlett",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Journal of Machine Learning Research",
                "volume": "6",
                "issue": "",
                "pages": "1817--1853",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rie Kubota Ando, Tong Zhang, and Peter Bartlett. 2005. A framework for learning predictive struc- tures from multiple tasks and unlabeled data. Jour- nal of Machine Learning Research, 6:1817-1853.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Applying alternating structure optimization to word sense disambiguation",
                "authors": [
                    {
                        "first": "Ando",
                        "middle": [],
                        "last": "Rie Kubota",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the Tenth Conference on Computational Natural Language Learning (CoNLL-X)",
                "volume": "",
                "issue": "",
                "pages": "77--84",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rie Kubota Ando. 2006. Applying alternating struc- ture optimization to word sense disambiguation. In Proceedings of the Tenth Conference on Compu- tational Natural Language Learning (CoNLL-X), pages 77-84, June.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Neural Networks for Pattern Recognition",
                "authors": [
                    {
                        "first": "M",
                        "middle": [],
                        "last": "Bishop",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "M. Bishop. 1995. Neural Networks for Pattern Recognition. Oxford University Press.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Subsequence kernels for relation extraction",
                "authors": [
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Razvan",
                        "suffix": ""
                    },
                    {
                        "first": "Raymond",
                        "middle": [
                            "J"
                        ],
                        "last": "Bunescu",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Mooney",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "NIPS 2005",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Razvan C. Bunescu and Raymond J. Mooney. 2005. Subsequence kernels for relation extraction. In NIPS 2005.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Comparative experiments on learning information extractors for proteins and their interactions",
                "authors": [
                    {
                        "first": "Yuk",
                        "middle": [
                            "Wah"
                        ],
                        "last": "Ramani",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Wong",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Artificial Intelligence in Medicine",
                "volume": "33",
                "issue": "2",
                "pages": "139--155",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ramani, and Yuk Wah Wong. 2005. Comparative experiments on learning information extractors for proteins and their interactions. Artificial Intelligence in Medicine, 33(2):139-155.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Discriminative training methods for hidden markov models: theory and experiments with perceptron algorithms",
                "authors": [
                    {
                        "first": "Michael",
                        "middle": [],
                        "last": "Collins",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "EMNLP 2002",
                "volume": "",
                "issue": "",
                "pages": "1--8",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Michael Collins. 2002. Discriminative training meth- ods for hidden markov models: theory and experi- ments with perceptron algorithms. In EMNLP 2002, pages 1-8.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Boosting for transfer learning",
                "authors": [
                    {
                        "first": "Wenyuan",
                        "middle": [],
                        "last": "Dai",
                        "suffix": ""
                    },
                    {
                        "first": "Qiang",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Gui-Rong",
                        "middle": [],
                        "last": "Xue",
                        "suffix": ""
                    },
                    {
                        "first": "Yong",
                        "middle": [],
                        "last": "Yu",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "ICML 2007",
                "volume": "",
                "issue": "",
                "pages": "193--200",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wenyuan Dai, Qiang Yang, Gui-Rong Xue, and Yong Yu. 2007. Boosting for transfer learning. In ICML 2007, pages 193-200.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Mining medline: abstracts, sentences, or phrases?",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Ding",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Berleant",
                        "suffix": ""
                    },
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Nettleton",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Wurtele",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Pacific Symposium on Biocomputing",
                "volume": "",
                "issue": "",
                "pages": "326--337",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Ding, D. Berleant, D. Nettleton, and E. Wurtele. 2002. Mining medline: abstracts, sentences, or phrases? Pacific Symposium on Biocomputing, pages 326-337.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Confidence-weighted linear classification",
                "authors": [
                    {
                        "first": "Mark",
                        "middle": [],
                        "last": "Dredze",
                        "suffix": ""
                    },
                    {
                        "first": "Koby",
                        "middle": [],
                        "last": "Crammer",
                        "suffix": ""
                    },
                    {
                        "first": "Fernando",
                        "middle": [],
                        "last": "Pereira",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "264--271",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Mark Dredze, Koby Crammer, and Fernando Pereira. 2008. Confidence-weighted linear classification. In ICML 2008, pages 264-271.",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Semi-supervised classification for extracting protein interaction sentences using dependency parsing",
                "authors": [
                    {
                        "first": "Gunes",
                        "middle": [],
                        "last": "Erkan",
                        "suffix": ""
                    },
                    {
                        "first": "Arzucan",
                        "middle": [],
                        "last": "Ozgur",
                        "suffix": ""
                    },
                    {
                        "first": "Dragomir",
                        "middle": [
                            "R"
                        ],
                        "last": "Radev",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gunes Erkan, Arzucan Ozgur, and Dragomir R. Radev. 2007. Semi-supervised classification for extract- ing protein interaction sentences using dependency parsing. In EMNLP 2007.",
                "links": null
            },
            "BIBREF13": {
                "ref_id": "b13",
                "title": "LIBLINEAR: A library for large linear classification",
                "authors": [
                    {
                        "first": "Kai-Wei",
                        "middle": [],
                        "last": "Rong-En Fan",
                        "suffix": ""
                    },
                    {
                        "first": "Cho-Jui",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Xiang-Rui",
                        "middle": [],
                        "last": "Hsieh",
                        "suffix": ""
                    },
                    {
                        "first": "Chih-Jen",
                        "middle": [],
                        "last": "Wang",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Journal of Machine Learning Research",
                "volume": "9",
                "issue": "",
                "pages": "1871--1874",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rong-En Fan, Kai-Wei Chang, Cho-Jui Hsieh, Xiang- Rui Wang, and Chih-Jen Lin. 2008. LIBLINEAR: A library for large linear classification. Journal of Machine Learning Research, 9:1871-1874.",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "Relex-relation extraction using dependency parse trees",
                "authors": [
                    {
                        "first": "Katrin",
                        "middle": [],
                        "last": "Fundel",
                        "suffix": ""
                    },
                    {
                        "first": "Robert",
                        "middle": [],
                        "last": "K\u00fcffner",
                        "suffix": ""
                    },
                    {
                        "first": "Ralf",
                        "middle": [],
                        "last": "Zimmer",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Bioinformatics",
                "volume": "23",
                "issue": "3",
                "pages": "365--371",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Katrin Fundel, Robert K\u00fcffner, and Ralf Zimmer. 2006. Relex-relation extraction using dependency parse trees. Bioinformatics, 23(3):365-371.",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "A dual coordinate descent method for large-scale linear SVM",
                "authors": [
                    {
                        "first": "Cho-Jui",
                        "middle": [],
                        "last": "Hsieh",
                        "suffix": ""
                    },
                    {
                        "first": "Kai-Wei",
                        "middle": [],
                        "last": "Chang",
                        "suffix": ""
                    },
                    {
                        "first": "Chih-Jen",
                        "middle": [],
                        "last": "Lin",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Sathiya",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Keerthi",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Sundararajan",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "408--415",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Cho-Jui Hsieh, Kai-Wei Chang, Chih-Jen Lin, S. Sathiya Keerthi, and S. Sundararajan. 2008. A dual coordinate descent method for large-scale lin- ear SVM. In ICML 2008, pages 408-415.",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "GENIA corpus -a semantically annotated corpus for bio-textmining",
                "authors": [
                    {
                        "first": "Jin-Dong",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Tomoko",
                        "middle": [],
                        "last": "Ohta",
                        "suffix": ""
                    },
                    {
                        "first": "Yuka",
                        "middle": [],
                        "last": "Tateisi",
                        "suffix": ""
                    },
                    {
                        "first": "Jun'ichi",
                        "middle": [],
                        "last": "Tsujii",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Bioinformatics",
                "volume": "19",
                "issue": "",
                "pages": "180--182",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jin-Dong Kim, Tomoko Ohta, Yuka Tateisi, and Jun'ichi Tsujii. 2003. GENIA corpus -a semanti- cally annotated corpus for bio-textmining. Bioinfor- matics, 19:i180-i182.",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Corpus annotation for mining biomedical events from literature",
                "authors": [
                    {
                        "first": "Jin-Dong",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Tomoko",
                        "middle": [],
                        "last": "Ohta",
                        "suffix": ""
                    },
                    {
                        "first": "Jun'ichi",
                        "middle": [],
                        "last": "Tsujii",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "BMC Bioinformatics",
                "volume": "9",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jin-Dong Kim, Tomoko Ohta, and Jun'ichi Tsujii. 2008a. Corpus annotation for mining biomedical events from literature. BMC Bioinformatics, 9:10.",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Kernel approaches for genic interaction extraction",
                "authors": [
                    {
                        "first": "Seonho",
                        "middle": [],
                        "last": "Kim",
                        "suffix": ""
                    },
                    {
                        "first": "Juntae",
                        "middle": [],
                        "last": "Yoon",
                        "suffix": ""
                    },
                    {
                        "first": "Jihoon",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Bioinformatics",
                "volume": "24",
                "issue": "1",
                "pages": "118--126",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Seonho Kim, Juntae Yoon, and Jihoon Yang. 2008b. Kernel approaches for genic interaction extraction. Bioinformatics, 24(1):118-126.",
                "links": null
            },
            "BIBREF19": {
                "ref_id": "b19",
                "title": "An evaluation of human protein-protein interaction data in the public domain",
                "authors": [
                    {
                        "first": "Balamurugan",
                        "middle": [],
                        "last": "Suresh Mathivanan",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Periaswamy",
                        "suffix": ""
                    },
                    {
                        "first": "Kumaran",
                        "middle": [],
                        "last": "Gandhi",
                        "suffix": ""
                    },
                    {
                        "first": "Shubha",
                        "middle": [],
                        "last": "Kandasamy",
                        "suffix": ""
                    },
                    {
                        "first": "Riaz",
                        "middle": [],
                        "last": "Suresh",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Mohmood",
                        "suffix": ""
                    },
                    {
                        "first": "Akhilesh",
                        "middle": [],
                        "last": "Yl Ramachandra",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Pandey",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "BMC Bioinformatics",
                "volume": "7",
                "issue": "5",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Suresh Mathivanan, Balamurugan Periaswamy, TKB Gandhi, Kumaran Kandasamy, Shubha Suresh, Riaz Mohmood, YL Ramachandra, and Akhilesh Pandey. 2006. An evaluation of human protein-protein inter- action data in the public domain. BMC Bioinformat- ics, 7 Suppl 5:S19.",
                "links": null
            },
            "BIBREF20": {
                "ref_id": "b20",
                "title": "Note on the sampling error of the difference between correlated proportions or percentages",
                "authors": [
                    {
                        "first": "Quinn",
                        "middle": [],
                        "last": "Mcnemar",
                        "suffix": ""
                    }
                ],
                "year": 1947,
                "venue": "Psychometrika",
                "volume": "12",
                "issue": "2",
                "pages": "153--157",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Quinn McNemar. 1947. Note on the sampling error of the difference between correlated proportions or percentages. Psychometrika, 12(2):153-157, June.",
                "links": null
            },
            "BIBREF21": {
                "ref_id": "b21",
                "title": "Combining multiple layers of syntactic information for proteinprotein interaction extraction",
                "authors": [
                    {
                        "first": "Makoto",
                        "middle": [],
                        "last": "Miwa",
                        "suffix": ""
                    },
                    {
                        "first": "Rune",
                        "middle": [],
                        "last": "Saetre",
                        "suffix": ""
                    },
                    {
                        "first": "Yusuke",
                        "middle": [],
                        "last": "Miyao",
                        "suffix": ""
                    },
                    {
                        "first": "Tomoko",
                        "middle": [],
                        "last": "Ohta",
                        "suffix": ""
                    },
                    {
                        "first": "Jun'ichi",
                        "middle": [],
                        "last": "Tsujii",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the Third International Symposium on Semantic Mining in Biomedicine (SMBM 2008)",
                "volume": "",
                "issue": "",
                "pages": "101--108",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Makoto Miwa, Rune Saetre, Yusuke Miyao, Tomoko Ohta, and Jun'ichi Tsujii. 2008. Combining mul- tiple layers of syntactic information for protein- protein interaction extraction. In Proceedings of the Third International Symposium on Semantic Mining in Biomedicine (SMBM 2008), pages 101-108.",
                "links": null
            },
            "BIBREF22": {
                "ref_id": "b22",
                "title": "Taskoriented evaluation of syntactic parsers and their representations",
                "authors": [
                    {
                        "first": "Yusuke",
                        "middle": [],
                        "last": "Miyao",
                        "suffix": ""
                    },
                    {
                        "first": "Rune",
                        "middle": [],
                        "last": "Saetre",
                        "suffix": ""
                    },
                    {
                        "first": "Kenji",
                        "middle": [],
                        "last": "Sagae",
                        "suffix": ""
                    },
                    {
                        "first": "Takuya",
                        "middle": [],
                        "last": "Matsuzaki",
                        "suffix": ""
                    },
                    {
                        "first": "Jun'ichi",
                        "middle": [],
                        "last": "Tsujii",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the 45th Meeting of the Association for Computational Linguistics (ACL'08:HLT)",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yusuke Miyao, Rune Saetre, Kenji Sagae, Takuya Matsuzaki, and Jun'ichi Tsujii. 2008. Task- oriented evaluation of syntactic parsers and their representations. In Proceedings of the 45th Meet- ing of the Association for Computational Linguistics (ACL'08:HLT).",
                "links": null
            },
            "BIBREF23": {
                "ref_id": "b23",
                "title": "Learning language in logicgenic interaction extraction challenge",
                "authors": [
                    {
                        "first": "Claire",
                        "middle": [],
                        "last": "N\u00e9dellec",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Proceedings of the LLL'05 Workshop",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Claire N\u00e9dellec. 2005. Learning language in logic - genic interaction extraction challenge. In Proceed- ings of the LLL'05 Workshop.",
                "links": null
            },
            "BIBREF24": {
                "ref_id": "b24",
                "title": "A survey on transfer learning",
                "authors": [
                    {
                        "first": "Qiang",
                        "middle": [],
                        "last": "Sinno Jialin Pan",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sinno Jialin Pan and Qiang Yang. 2008. A survey on transfer learning. Technical Report HKUST-CS08- 08, Department of Computer Science and Engineer- ing, Hong Kong University of Science and Technol- ogy, Hong Kong, China, November.",
                "links": null
            },
            "BIBREF25": {
                "ref_id": "b25",
                "title": "BioInfer: A corpus for information extraction in the biomedical domain",
                "authors": [
                    {
                        "first": "Sampo",
                        "middle": [],
                        "last": "Pyysalo",
                        "suffix": ""
                    },
                    {
                        "first": "Filip",
                        "middle": [],
                        "last": "Ginter",
                        "suffix": ""
                    },
                    {
                        "first": "Juho",
                        "middle": [],
                        "last": "Heimonen",
                        "suffix": ""
                    },
                    {
                        "first": "Jari",
                        "middle": [],
                        "last": "Bj\u00f6rne",
                        "suffix": ""
                    },
                    {
                        "first": "Jorma",
                        "middle": [],
                        "last": "Boberg",
                        "suffix": ""
                    },
                    {
                        "first": "Jouni",
                        "middle": [],
                        "last": "J\u00e4rvinen",
                        "suffix": ""
                    },
                    {
                        "first": "Tapio",
                        "middle": [],
                        "last": "Salakoski",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "BMC Bioinformatics",
                "volume": "8",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sampo Pyysalo, Filip Ginter, Juho Heimonen, Jari Bj\u00f6rne, Jorma Boberg, Jouni J\u00e4rvinen, and Tapio Salakoski. 2007. BioInfer: A corpus for infor- mation extraction in the biomedical domain. BMC Bioinformatics, 8:50.",
                "links": null
            },
            "BIBREF26": {
                "ref_id": "b26",
                "title": "Comparative analysis of five protein-protein interaction corpora",
                "authors": [
                    {
                        "first": "Sampo",
                        "middle": [],
                        "last": "Pyysalo",
                        "suffix": ""
                    },
                    {
                        "first": "Antti",
                        "middle": [],
                        "last": "Airola",
                        "suffix": ""
                    },
                    {
                        "first": "Juho",
                        "middle": [],
                        "last": "Heimonen",
                        "suffix": ""
                    },
                    {
                        "first": "Jari",
                        "middle": [],
                        "last": "Bj\u00f6rne",
                        "suffix": ""
                    },
                    {
                        "first": "Filip",
                        "middle": [],
                        "last": "Ginter",
                        "suffix": ""
                    },
                    {
                        "first": "Tapio",
                        "middle": [],
                        "last": "Salakoski",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "BMC Bioinformatics",
                "volume": "9",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Sampo Pyysalo, Antti Airola, Juho Heimonen, Jari Bj\u00f6rne, Filip Ginter, and Tapio Salakoski. 2008. Comparative analysis of five protein-protein inter- action corpora. In BMC Bioinformatics, volume 9(Suppl 3), page S6.",
                "links": null
            },
            "BIBREF27": {
                "ref_id": "b27",
                "title": "Soft margins for adaboost",
                "authors": [
                    {
                        "first": "Gunnar",
                        "middle": [],
                        "last": "Ratsch",
                        "suffix": ""
                    },
                    {
                        "first": "Takashi",
                        "middle": [],
                        "last": "Onoda",
                        "suffix": ""
                    },
                    {
                        "first": "Klaus-Robert",
                        "middle": [],
                        "last": "Muller",
                        "suffix": ""
                    }
                ],
                "year": 2001,
                "venue": "Machine Learning",
                "volume": "42",
                "issue": "",
                "pages": "287--320",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gunnar Ratsch, Takashi Onoda, and Klaus-Robert Muller. 2001. Soft margins for adaboost. Machine Learning, 42(3):287-320.",
                "links": null
            },
            "BIBREF28": {
                "ref_id": "b28",
                "title": "Syntactic features for protein-protein interaction extraction",
                "authors": [
                    {
                        "first": "Rune",
                        "middle": [],
                        "last": "Saetre",
                        "suffix": ""
                    },
                    {
                        "first": "Kenji",
                        "middle": [],
                        "last": "Sagae",
                        "suffix": ""
                    },
                    {
                        "first": "Jun'ichi",
                        "middle": [],
                        "last": "Tsujii",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "LBM 2007 short papers",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rune Saetre, Kenji Sagae, and Jun'ichi Tsujii. 2007. Syntactic features for protein-protein interaction ex- traction. In LBM 2007 short papers.",
                "links": null
            },
            "BIBREF29": {
                "ref_id": "b29",
                "title": "An empirical analysis of domain adaptation algorithms for genomic sequence analysis",
                "authors": [
                    {
                        "first": "Gabriele",
                        "middle": [],
                        "last": "Schweikert",
                        "suffix": ""
                    },
                    {
                        "first": "Christian",
                        "middle": [],
                        "last": "Widmer",
                        "suffix": ""
                    },
                    {
                        "first": "Bernhard",
                        "middle": [],
                        "last": "Sch\u00f6lkopf",
                        "suffix": ""
                    },
                    {
                        "first": "Gunnar",
                        "middle": [],
                        "last": "R\u00e4tsch",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "NIPS",
                "volume": "",
                "issue": "",
                "pages": "1433--1440",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Gabriele Schweikert, Christian Widmer, Bernhard Sch\u00f6lkopf, and Gunnar R\u00e4tsch. 2008. An empir- ical analysis of domain adaptation algorithms for genomic sequence analysis. In NIPS, pages 1433- 1440.",
                "links": null
            },
            "BIBREF30": {
                "ref_id": "b30",
                "title": "Cross-domain video concept detection using adaptive SVMs",
                "authors": [
                    {
                        "first": "Jun",
                        "middle": [],
                        "last": "Yang",
                        "suffix": ""
                    },
                    {
                        "first": "Rong",
                        "middle": [],
                        "last": "Yan",
                        "suffix": ""
                    },
                    {
                        "first": "Alexander",
                        "middle": [
                            "G"
                        ],
                        "last": "Hauptmann",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "MULTIMEDIA '07: Proceedings of the 15th international conference on Multimedia",
                "volume": "",
                "issue": "",
                "pages": "188--197",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jun Yang, Rong Yan, and Alexander G. Hauptmann. 2007. Cross-domain video concept detection using adaptive SVMs. In MULTIMEDIA '07: Proceed- ings of the 15th international conference on Multi- media, pages 188-197.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "text": "A sentence including an interacting protein pair (p1, p2",
                "num": null,
                "type_str": "figure",
                "uris": null
            },
            "FIGREF1": {
                "text": "Extraction of a feature vector from the target sentence of proteins in a sentence is interacting or not.Figure 2shows an example of a sentence in which the given pair (p1 and p2) actually interacts.",
                "num": null,
                "type_str": "figure",
                "uris": null
            },
            "FIGREF2": {
                "text": "summarizes the way in which the feature vector is constructed. The system extracts Bag-of-Words (BOW), shortest path (SP), and graph features from the output of two parsers. The PROT M:1, and M:1, interact M:1, multiple M:1, of M:1, protein M:1, subunit M:1, with M:2, protein A:1 Bag-of-Words features of the pair in Figure 2 with their positions (B:Before, M:in the Middle of, A:After) and frequencies.",
                "num": null,
                "type_str": "figure",
                "uris": null
            },
            "FIGREF3": {
                "text": "Vertex walks, edge walks in the upper shortest path between the proteins in the parse tree by KSDEP. The walks and their subsets are used as the shortest path features of the pair inFigure 2.",
                "num": null,
                "type_str": "figure",
                "uris": null
            },
            "FIGREF4": {
                "text": "Figure 5illustrates the shortest path between the pair in Figure 2, and its v-walks and e-walks extracted from the shortest path in the parse tree by KSDEP. A v-walk includes two lemmas and their link, while",
                "num": null,
                "type_str": "figure",
                "uris": null
            },
            "TABREF1": {
                "text": "The sizes of used PPI corpora. A:AIMed, B:BioInfer, H:HPRD50, I:IEPA, and L:LLL.",
                "num": null,
                "html": null,
                "content": "<table><tr><td/><td/><td/><td colspan=\"2\">A</td><td/><td>B</td><td/><td>H</td><td/><td>I</td><td>L</td></tr><tr><td/><td colspan=\"10\">positive 1,000 2,534 163 335 164</td></tr><tr><td/><td>all</td><td/><td colspan=\"8\">5,834 9,653 433 817 330</td></tr><tr><td>1 0</td><td>0</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>9</td><td>0</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>8</td><td>0</td><td/><td/><td/><td/><td/><td/><td/><td/><td>A B i I m e o I n f d e ( F r ( ) F )</td></tr><tr><td>7</td><td>0</td><td/><td/><td/><td/><td/><td/><td/><td/><td>A B i I m e o I n f d e ( A U C ) r ( A U C )</td></tr><tr><td>6</td><td>0</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>5</td><td>0</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>0</td><td>2 0</td><td>4 % 0</td><td>e x a m</td><td>p l</td><td>6 e s 0</td><td>8</td><td>0</td><td>1</td><td>0 0</td></tr></table>",
                "type_str": "table"
            },
            "TABREF5": {
                "text": "Comparison of methods on multiple corpora. Our feature vector without graph features is used. The source corpora with the best F-scores are reported for aSVM, TrAdaBoost, and SVM-CW.",
                "num": null,
                "html": null,
                "content": "<table><tr><td/><td/><td/><td colspan=\"2\">F-score</td><td/><td/><td/><td/><td>AUC</td><td/><td/></tr><tr><td/><td>A</td><td>B</td><td>H</td><td>I</td><td>L</td><td>all</td><td>A</td><td>B</td><td>H</td><td>I</td><td>L</td><td>all</td></tr><tr><td colspan=\"3\">A (64.2) 64.0</td><td>64.7</td><td>65.2</td><td colspan=\"4\">63.7 64.2 (89.1) 89.5</td><td>89.2</td><td>89.3</td><td colspan=\"2\">89.0 89.4</td></tr><tr><td colspan=\"4\">B 67.9 (67.6) 67.9</td><td>67.9</td><td colspan=\"5\">67.7 68.3 86.2 (86.1) 86.2</td><td>86.3</td><td colspan=\"2\">86.2 86.4</td></tr><tr><td colspan=\"2\">H 71.3</td><td colspan=\"3\">71.2 (69.7) 74.1</td><td colspan=\"3\">70.8 74.9 84.7</td><td colspan=\"3\">85.0 (82.8) 85.0</td><td colspan=\"2\">83.4 87.9</td></tr><tr><td>I</td><td>74.4</td><td>75.6</td><td colspan=\"5\">73.7 (74.4) 74.4 76.6 86.7</td><td>87.1</td><td colspan=\"4\">85.4 (85.6) 86.9 87.8</td></tr><tr><td>L</td><td>83.2</td><td>85.9</td><td>82.0</td><td colspan=\"4\">86.7 (80.5) 84.1 86.3</td><td>87.1</td><td>87.4</td><td colspan=\"3\">90.8 (86.0) 86.2</td></tr></table>",
                "type_str": "table"
            }
        }
    }
}