File size: 51,435 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
{
    "paper_id": "R09-1024",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T15:01:00.383617Z"
    },
    "title": "Learning to Identify Educational Materials",
    "authors": [
        {
            "first": "Samer",
            "middle": [],
            "last": "Hassan",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of North Texas",
                "location": {}
            },
            "email": ""
        },
        {
            "first": "Rada",
            "middle": [],
            "last": "Mihalcea",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "University of North Texas",
                "location": {}
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "In this paper, we explore the task of automatically identifying educational materials, by classifying documents with respect to their educative value. Through experiments carried out on a data set of manually annotated documents, we show that the generally accepted notion of a learning object's \"educativeness\" is indeed a property that can be reliably assigned through automatic classification.",
    "pdf_parse": {
        "paper_id": "R09-1024",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "In this paper, we explore the task of automatically identifying educational materials, by classifying documents with respect to their educative value. Through experiments carried out on a data set of manually annotated documents, we show that the generally accepted notion of a learning object's \"educativeness\" is indeed a property that can be reliably assigned through automatic classification.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "With the rapid growth of the amount of information available online and elsewhere, it becomes increasingly difficult to identify documents that satisfy the user needs. Current search engines target broad coverage of information, at the cost of providing limited support for well defined verticals.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In particular, an increasingly large number of users, consisting primarily of students, instructors and selftaught learners, are often seeking educational materials online, to use as standalone instructional materials or to supplement existing class resources. The typical solution is to either refer to existing collections of learning materials, which often lack breadth of coverage, or to search the Web using one of the current search engines, which frequently lead to many irrelevant results. For example, as shown later in Section 3, from the top 50 documents returned by a search performed on a major search engine 1 for the query \"tree data structure,\" only four were found to be strongly educative, while as many as 29 documents were found to be non-educative.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In this paper, we address the task of automatically identifying educational materials. We formulate the task as a text categorization problem, and try to automatically classify the \"educativeness\" of a document (defined as a property that reflects the educative value of a document). Through annotation experiments carried out on a data set of materials from the domain of computer science, we show that the educativeness of a document is a property that can be reliably assigned by human judges. We also identify several features characteristic to educational resources, which can be used to identify the educativeness of a document. We perform a number of classification experiments, and show that the document educativeness can be learned and automatically assigned.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "A learning object is formally defined as \"any entity, digital or non-digital, that may be used for learning, education or training\" [2] , or \"any digital resource that can be reused to support learning\" [12] .",
                "cite_spans": [
                    {
                        "start": 132,
                        "end": 135,
                        "text": "[2]",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 203,
                        "end": 207,
                        "text": "[12]",
                        "ref_id": "BIBREF11"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Background",
                "sec_num": "2"
            },
            {
                "text": "The idea that a document can have an educative property is widely accepted in the growing body of work dedicated to learning objects. Learning object repositories (e.g., [6, 8] ) target improved access to learning materials through \"sharing and reuse,\" by providing a common interface to entire collections of learning materials that can be shared among students and instructors and can be reused across courses and disciplines. These definitions are representative for the notion of \"educativeness\" as used in this paper.",
                "cite_spans": [
                    {
                        "start": 170,
                        "end": 173,
                        "text": "[6,",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 174,
                        "end": 176,
                        "text": "8]",
                        "ref_id": "BIBREF7"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Background",
                "sec_num": "2"
            },
            {
                "text": "While there has been a large body of work focused toward Learning Object Metadata harnessing [7, 4, 1], we are not aware of any work that has tried to harness the power of the Web as an educational resource through the automatic identification of learning assets on the Web. The work closest to ours is perhaps [9] , where the authors addressed the problem of finding educational resources on the Web. However, the focus of their work was limited to metadata extraction for a limited set of fine grained properties. Instead, in this paper, we introduce a method to automatically annotate the educativeness property of a document, which can be used to assist learners in their search for educational materials.",
                "cite_spans": [
                    {
                        "start": 311,
                        "end": 314,
                        "text": "[9]",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Background",
                "sec_num": "2"
            },
            {
                "text": "It is important to note that the classification of the educativeness of a document cannot be modeled as a genre classification task. While recognizing the educativeness of a document is relatively easy to do with accomplished readers, different educational materials can have major stylistic inconsistencies, which invalidate their membership to a unified genre [3] . For example, a diagram, textbook, and a blog could all serve as useful and educative resources despite their obvious stylistic differences.",
                "cite_spans": [
                    {
                        "start": 362,
                        "end": 365,
                        "text": "[3]",
                        "ref_id": "BIBREF2"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Background",
                "sec_num": "2"
            },
            {
                "text": "What is an educational material? The purpose of educational materials is primarily decided by the author or the presenter of the resource, who furthermore decides the target audience and the delivery style (e.g., textbooks, presentation, diagram). While the purpose of the resource is a property that is mainly determined by its author, the strength of the educative resource (\"educativeness\") is a property evaluated cumulatively by the target audience of the resource (e.g., students or educational experts) . Hence, in the construction of our data set and in the evaluations we run, we focus on the educativeness property of a learning resource as determined by the agreement of their potential users (students). Educational materials can be located in a variety of sources and formats, including lectures, tutorials, online books, blog articles, publications, even technical forums or expert networks. Most of these learning objects typically include several of the following components: definitions, examples, questions and answers, diagrams, and illustrations.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Building A Data Set for the Classification of Educational Materials",
                "sec_num": "3"
            },
            {
                "text": "In order to build a data set for the classification of educational materials, we mimic a hypothetical learner who tries to locate and identify learning assets using current online resources. We use a typical search scenario, which involves the use of a search engine with a disambiguated query to identify candidate materials, followed by a filtering step that selects only those materials that have educational relevance.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Building A Data Set for the Classification of Educational Materials",
                "sec_num": "3"
            },
            {
                "text": "We collect a data set covering the domain of computer science. We select fourteen topics frequently addressed in data structures and algorithms courses, as shown in Table 2 . Starting with each of the fourteen topics, a query is constructed and run against the Google search engine, and the top 60 ranked search results are collected. 2 Note that the meaning of some terms can be ambiguous, e.g., \"tree\" or \"list,\" and thus we explicitly disambiguate the query by adding the phrase \"data structure.\" By performing this explicit disambiguation, we can focus on the educativeness property of the documents returned by the search, rather than on the differences that could arise from ambiguities of meaning.",
                "cite_spans": [
                    {
                        "start": 335,
                        "end": 336,
                        "text": "2",
                        "ref_id": "BIBREF1"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 165,
                        "end": 172,
                        "text": "Table 2",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Building A Data Set for the Classification of Educational Materials",
                "sec_num": "3"
            },
            {
                "text": "We define a set of features largely based on the properties associated with learning objects, as defined in standards such as IEEE LOM [2] . Some of the features are also motivated by previous work on educational metadata [11] . The following features are associated with each document in the data set.",
                "cite_spans": [
                    {
                        "start": 135,
                        "end": 138,
                        "text": "[2]",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 222,
                        "end": 226,
                        "text": "[11]",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Properties of Educational Materials",
                "sec_num": "3.1"
            },
            {
                "text": "To be able to capture the educativeness of a resource, the annotators had to score each page on its overall educative value. This feature serves as the major class of the documents in the data set. The annotators were instructed to evaluate the resource as a necessary asset for a student to understand the topic, and score each document on a four point scale ranging from \"noneducative\" to \"strongly-educative.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Educativeness",
                "sec_num": null
            },
            {
                "text": "We want to measure how human-assigned relevance can contribute to our task, and see if an accurate (manual) measure of relevance can result in a better identification of learning objects, as compared to the search engine ranking. We measure relevance on a four point scale ranging from \"non-relevant\" to \"very relevant.\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Relevance",
                "sec_num": null
            },
            {
                "text": "The content category is a feature that classifies the type of content found on the target page. We assume that the typical content of a learning object can be categorized into one or more of the following types: Definition: The content presents a textual definition of a concept or any of its associated properties. Example/Use: The content presents examples that help clarify a concept, demonstrative use of a concept, or the use of operations in that concept. (e.g., the queue data structure push and pop operations) Questions & Answers: The content presents a question and answer dialogue, as usually found in technical forums and sometime in blog articles. Illustration: The content presents an illustration of a concept or a process, either through the use of images, or through diagrams.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Content Categories",
                "sec_num": null
            },
            {
                "text": "Other: This group contains all the other types that do not fit in the previous categories.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Content Categories",
                "sec_num": null
            },
            {
                "text": "One of the interesting properties of the learning asset is its source. Under the assumption that the type of the resource can contribute to the document educativeness, the annotators were instructed to choose all the possible types that apply from a pre-compiled list. The list was generated by observing and inspecting the collection of retrieved documents. These types are not mutually exclusive.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resource Type",
                "sec_num": null
            },
            {
                "text": "Class webpage: A typical class home page where the teacher would provide lecture notes, tips, quizzes and answers for the class homework. Encyclopedia: A resource for educative materials, representing semi-structured or fully structured knowledge contributed by experts in the field. Blog: Web log or blog represents an online personal journal. It varies in format and purpose and it is an increasing popular online form of self-expression.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resource Type",
                "sec_num": null
            },
            {
                "text": "Mailing list/forums: It is a typical example of expert network where users pose their questions to an expert (or group of experts) in the field and receive one or more answers. Usually such content is very technical but not always useful.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resource Type",
                "sec_num": null
            },
            {
                "text": "Online book: This category represents electronic books in an online format (e.g., HTML, PDF). Presentation: A demonstrative material that consists of a set of slides or pages, representing the main points to be addressed with respect to a topic. Publication: This group includes scientific publications, such as journal articles, conference proceedings, article abstracts, and patent descriptions. How-To article: This source type addresses the use of a specific concept on a step by step basis.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resource Type",
                "sec_num": null
            },
            {
                "text": "Reference manual: A technical reference or manual, which explains the use and the inner workings of a concept (e.g., Java language documentations).",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resource Type",
                "sec_num": null
            },
            {
                "text": "Other: This category includes all other content (e.g. product catalogs, company homepages)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resource Type",
                "sec_num": null
            },
            {
                "text": "Learning objects are very diverse and are subject to the judgment of the learner. An expert in the field needs little introduction to the topic, and may require a high level of technical insight. Instead, the same information might seem non-educative and irrelevant from the perspective of a novice user, who seeks basic fundamentals. To address this problem, we asked the annotators to indicate their expertise in each of the selected topics on a four point scale.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Expertise",
                "sec_num": null
            },
            {
                "text": "Taken together, all the features defined above are referred to as \"user features,\" and are listed in Table  1 . In addition to these features, for each document in the data set we also collect its search engine ranking and its document type (ppt, pdf, html, doc, etc.). We also calculate the hubness of each page as a ratio of its hyper-linked contents to its original content. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 101,
                        "end": 109,
                        "text": "Table  1",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Final Set of Features",
                "sec_num": "3.2"
            },
            {
                "text": "Two judges individually annotated the collected documents based on a set of annotation guidelines. The annotators were required to identify the value associated with all the document features described above, along with the educativeness property of a document. The annotators were instructed to evaluate the resource from a college student perspective, therefore discarding highly technical and specific resources as non-educative or marginally-educative. We measure the inter-annotator agreement by calculating the kappa statistic for the annotations made by the two human annotators. The inter-annotator agreement and the kappa statistic for all the features are shown in Figure 1 .",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 675,
                        "end": 683,
                        "text": "Figure 1",
                        "ref_id": "FIGREF0"
                    }
                ],
                "eq_spans": [],
                "section": "Agreement Study",
                "sec_num": "3.3"
            },
            {
                "text": "The final data set is created by asking a third annotator to arbitrate the disagreements among the first two annotators. The final distribution across the educativeness class labels is shown in Table 2 . As seen in the table, the distribution across educative and noneducative classes is relatively balanced with a few exceptions. Topics such as \"queue\" and \"tree\" tend to have more non-educative pages, unlike topics such as \"binary search,\" which tend to have more educative pages. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 194,
                        "end": 201,
                        "text": "Table 2",
                        "ref_id": "TABREF3"
                    }
                ],
                "eq_spans": [],
                "section": "Agreement Study",
                "sec_num": "3.3"
            },
            {
                "text": "Using the data set described in the previous section, we experiment with automatic classifiers to annotate the educativeness of a given document. Through these evaluations, we measure the ability of a system to automatically detect and classify documents according to their educative value.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "The four-point scale used for the educativeness annotation allows us to perform both a fine grained and a coarse grained evaluation. In the fine grained evaluation, all four dimensions are considered, and thus we run a four-way classification. In the coarse grained evaluation, we combine the non-educative and marginally educative documents into one class (noneducative), and the educative and strongly educative pages into another class (educative), and run a twoway classification. All the evaluations are conducted using a ten-fold cross validation.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "Through our experiments, we seek answers to the following questions:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "1. Can the content of a document be used to classify its educativeness? We evaluate the use of the doc-ument content to learn and detect its educativeness. The content is used to construct a feature representation of each document. The terms appearing in the learning objects serve as features in the learning algorithm, with a weight indicating their frequency in the learning object.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "2. Are the user-features useful for the classification of a document educativeness? We evaluate the selected user features as possible dimensions to learn and detect the educativeness of target examples. We use all the user features summarized in Table 1 to construct a feature vector representation for each learning asset. Since these features were manually assigned by the annotators, these annotations serve as an upper bound on the accuracy that can be achieved by using such features.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 247,
                        "end": 254,
                        "text": "Table 1",
                        "ref_id": "TABREF1"
                    }
                ],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "3. Can the content of a document be used to automatically predict the user-features? We run an evaluation where each of the selected user-features serves as its own class. The learning assets in which this feature has been selected by the annotators serve as positive examples, while the documents in which the feature was not encountered serve as negative examples. The content of the documents is used to build the feature vectors. The examples are then used to train a classifier to classify each of the features automatically.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "4. Can the automatically predicted user-features be used to learn and detect the educativeness of a document? Finally, given the set of classifiers generated in the previous experiment, we use their output to construct a machine weighted user-feature representation of the given document. This evaluation is similar to the one relying on manually assigned user-features. However, instead of using the user annotations, we use the output automatically predicted by the classifiers.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "For the experiments, we used two classifiers: Na\u00efve Bayes [5] and SVM [10] , selected based on their performance and diversity of learning methodologies.",
                "cite_spans": [
                    {
                        "start": 58,
                        "end": 61,
                        "text": "[5]",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 70,
                        "end": 74,
                        "text": "[10]",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Experiments",
                "sec_num": "4"
            },
            {
                "text": "We run a first experiment where we use the content of the documents, with minimal pre-processing (tokenization, stopword removal), and classify them with respect to the fine-grained and coarse-grained educativeness class. We use a 10-fold cross validation on the entire data set. The rows labeled with \"document content\" in Tables 3 show the results of this experiment. To answer the first question, these experiments show that the use of raw content is useful and can be effectively used to classify the educativeness of a document. In fact, compared to the baseline of selecting the most common class across all the documents, the content-based classification results in a 22-23% absolute increase in F-measure.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "5"
            },
            {
                "text": "Next, we use the manual annotations for the userfeatures to classify the educativeness of a document. The results obtained in this experiment are shown in Table 3 in the rows labeled with \"user-features (manual).\" The results are clearly superior, which answers the second question and suggests the usefulness of these features for the classification of educativeness. Note that these results represent an upper bound for our evaluations, since they rely on manually annotated features. Table 3 : Classification results",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 155,
                        "end": 162,
                        "text": "Table 3",
                        "ref_id": null
                    },
                    {
                        "start": 487,
                        "end": 494,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "5"
            },
            {
                "text": "Since the user-features seem to exhibit the best performance, next we evaluate the ability of automatically labeling these features using the content of the documents. The accuracy of the automatic classification of the user-features is shown in Figure 2 . Both SVM and Naive Bayes seem to be able to label these features with relatively high accuracy. The lowest performance is achieved for Relevance (50-56% F-measure) and the highest for IsEncyclopedia (86-95% F-measure). This experiment provides an answer to the third question: all the user-features that proved useful for the classification of educativeness can be predicted based on the document content. Finally, we answer the fourth question by running an experiment where the automatically predicted userfeatures are used as input to a classifier to annotate the educativeness of a document. The results obtained in this experiment are shown in Table 3 , under the rows labeled \"user-features (predicted).\" The performance obtained by this classifier shows slight advantage (1-5% absolute increase in F-measure) over the one obtained by using the raw content alone. This indicates that a prediction of high accuracy might help in closing the gap with the upper-bound obtained with the manually annotated user-features. This result can be the basis for future improvements, by seeking improvements in the classification of the individual features prediction (e.g., by using syntactic or semantic features in addition to lexical features).",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 246,
                        "end": 254,
                        "text": "Figure 2",
                        "ref_id": "FIGREF1"
                    },
                    {
                        "start": 906,
                        "end": 913,
                        "text": "Table 3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Features",
                "sec_num": null
            },
            {
                "text": "Based on our experiments, we found that the educativeness of a document is a property that can be automatically identified. Not surprisingly, the classification with respect to a set of coarse-grained classes is significantly higher than the fine-grained classification. In terms of features, the raw content of a document was found useful, as were other properties associated with a document (referred to as \"user-features\").",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Discussion",
                "sec_num": "6"
            },
            {
                "text": "To evaluate how each of the user-features contribute to the accuracy of the classification, we measured the information gain associated with each feature based on the manual annotations. Figure 3 shows the feature weights. Not surprisingly, the content categories (e.g., HasDefinition, HasExample, HasIllustration) score the highest, indicating their significant discriminative power. Interestingly, the Relevance feature has a higher discriminative power than the Rank feature, which indicates that the relevance of a document might be a good feature to consider when modeling its educativeness. Other intuitive features such as resource types (e.g., IsHowTo, IsPresentation) seem to also contribute to the classification. Note however that the degree of their contribution might be affected by the implicit dependency on content categories (e.g., pages classified as IsEncyclopedia often include definitions, which also activate the HasDefinition feature). ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 187,
                        "end": 195,
                        "text": "Figure 3",
                        "ref_id": "FIGREF2"
                    }
                ],
                "eq_spans": [],
                "section": "Discussion",
                "sec_num": "6"
            },
            {
                "text": "In this paper, we addressed the task of automatically identifying learning materials. We constructed a data set by manually annotating the educativeness of the documents retrieved for fourteen topics in computer science. An annotation experiment carried out on this data set showed that the educativeness of a document is a property that can be reliably assigned by human judges. Moreover, through a number of classification experiments, we showed that the educativeness property can also be automatically assigned, with up to 23% absolute increase in F-measure as compared to the most common class baseline.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "7"
            },
            {
                "text": "Through our experiments, we identified several promising lines for future research. First, we plan to explore ways of improving the classification accuracy for the individual user-features, as well as ways of combining them with the features extracted from the content of a document, in order to improve the overall accuracy of the classification of educativeness. Second, we plan to carry out larger-scale experiments to explore the portability across different domains.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "7"
            },
            {
                "text": "The data set introduced in the paper can be downloaded from http://lit.csci.unt.edu/index.php/Downloads",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Conclusion",
                "sec_num": "7"
            },
            {
                "text": "Throughout our experiments, we conduct our searches using the Google search engine.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "From the top 60 documents, some had to be removed prior to any further processing, because they were either unreachable or they contained non-English characters.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "The authors are grateful to Carmen Banea and Ravi Sinha for their help with the data annotations.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Metadata extraction and harvesting",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Greenberg",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Journal of Library Metadata",
                "volume": "6",
                "issue": "4",
                "pages": "59--82",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Greenberg. Metadata extraction and harvesting. Journal of Library Metadata, 6(4):59-82, 2004.",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Draft standard for learning technology -learning object metadata -iso/iec 11404",
                "authors": [
                    {
                        "first": "W",
                        "middle": [],
                        "last": "Hodgins",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Duval",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "W. Hodgins and E. Duval. Draft standard for learning tech- nology -learning object metadata -iso/iec 11404. Technical report, 2002.",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "The wheres and whyfores for studying textual genre computationally",
                "authors": [
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Karlgren",
                        "suffix": ""
                    }
                ],
                "year": 2004,
                "venue": "Proceedings of the AAAI Fall Symposium of Style and Meaning in Language, Art and Music",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "J. Karlgren. The wheres and whyfores for studying textual genre computationally. In In Proceedings of the AAAI Fall Symposium of Style and Meaning in Language, Art and Music., Washington D.C., 2004.",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Categorizing learning objects based on wikipedia as substitute corpus",
                "authors": [
                    {
                        "first": "",
                        "middle": [],
                        "last": "Marek",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Marek. Categorizing learning objects based on wikipedia as substitute corpus.",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "A comparison of event models for Naive Bayes text classification",
                "authors": [
                    {
                        "first": "A",
                        "middle": [],
                        "last": "Mccallum",
                        "suffix": ""
                    },
                    {
                        "first": "K",
                        "middle": [],
                        "last": "Nigam",
                        "suffix": ""
                    }
                ],
                "year": 1998,
                "venue": "Proceedings of AAAI Workshop on Learning for Text Categorization",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "A. McCallum and K. Nigam. A comparison of event models for Naive Bayes text classification. In Proceedings of AAAI Workshop on Learning for Text Categorization, 1998.",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Reusable learning objects: a survey of LOM-based repositories",
                "authors": [
                    {
                        "first": "F",
                        "middle": [],
                        "last": "Neven",
                        "suffix": ""
                    },
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Duval",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the ACM International Conference on Multimedia",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "F. Neven and E. Duval. Reusable learning objects: a survey of LOM-based repositories. In Proceedings of the ACM International Conference on Multimedia, France, 2002.",
                "links": null
            },
            "BIBREF6": {
                "ref_id": "b6",
                "title": "Strategies for automatic lom metadata generating in a web-based cscl tool",
                "authors": [
                    {
                        "first": "L",
                        "middle": [
                            "T E"
                        ],
                        "last": "Pansanato",
                        "suffix": ""
                    },
                    {
                        "first": "R",
                        "middle": [
                            "P M"
                        ],
                        "last": "Fortes",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "WebMedia '05: Proceedings of the 11th Brazilian Symposium on Multimedia and the web",
                "volume": "",
                "issue": "",
                "pages": "1--8",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "L. T. E. Pansanato and R. P. M. Fortes. Strategies for automatic lom metadata generating in a web-based cscl tool. In WebMedia '05: Proceedings of the 11th Brazilian Symposium on Multimedia and the web, pages 1-8, New York, NY, USA, 2005. ACM.",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Learning objects, learning object repositories and learning theory: Preliminary best practices for online courses",
                "authors": [
                    {
                        "first": "S",
                        "middle": [
                            "Smith"
                        ],
                        "last": "Nash",
                        "suffix": ""
                    }
                ],
                "year": 2005,
                "venue": "Interdisciplinary Journal of Knowledge and Learning Objects",
                "volume": "1",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "S. Smith Nash. Learning objects, learning object reposi- tories and learning theory: Preliminary best practices for online courses. Interdisciplinary Journal of Knowledge and Learning Objects, 1, 2005.",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "Finding educational resources on the web: Exploiting automatic extraction of metadata",
                "authors": [
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Thompson",
                        "suffix": ""
                    },
                    {
                        "first": "J",
                        "middle": [],
                        "last": "Smarr",
                        "suffix": ""
                    },
                    {
                        "first": "H",
                        "middle": [],
                        "last": "Nguyen",
                        "suffix": ""
                    },
                    {
                        "first": "C",
                        "middle": [],
                        "last": "Manning",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "ECML Workshop on Adaptive Text Extraction and Mining",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "C. Thompson, J. Smarr, H. Nguyen, and C. Manning. Finding educational resources on the web: Exploiting au- tomatic extraction of metadata. In ECML Workshop on Adaptive Text Extraction and Mining, 2003.",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "The Nature of Statistical Learning Theory",
                "authors": [
                    {
                        "first": "V",
                        "middle": [],
                        "last": "Vapnik",
                        "suffix": ""
                    }
                ],
                "year": 1995,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "V. Vapnik. The Nature of Statistical Learning Theory. Springer, New York, 1995.",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Creating glossaries using pattern-based and machine learning techniques",
                "authors": [
                    {
                        "first": "E",
                        "middle": [],
                        "last": "Westerhout",
                        "suffix": ""
                    },
                    {
                        "first": "P",
                        "middle": [],
                        "last": "Monachesi",
                        "suffix": ""
                    }
                ],
                "year": 2008,
                "venue": "Proceedings of the Sixth International Language Resources and Evaluation",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "E. Westerhout and P. Monachesi. Creating glossaries using pattern-based and machine learning techniques. In Pro- ceedings of the Sixth International Language Resources and Evaluation, Marrakech, Morocco, 2008.",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Learning Object Design and Sequencing Theory",
                "authors": [
                    {
                        "first": "D",
                        "middle": [],
                        "last": "Wiley",
                        "suffix": ""
                    }
                ],
                "year": 2000,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "D. Wiley. Learning Object Design and Sequencing Theory. PhD thesis, Department of Instructional Psychology and Technology Brigham Young University., 2000.",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF0": {
                "text": "Kappa statistic and inter-annotator agreement",
                "type_str": "figure",
                "uris": null,
                "num": null
            },
            "FIGREF1": {
                "text": "Classification results for user-features",
                "type_str": "figure",
                "uris": null,
                "num": null
            },
            "FIGREF2": {
                "text": "Information gain for user-features",
                "type_str": "figure",
                "uris": null,
                "num": null
            },
            "TABREF1": {
                "html": null,
                "num": null,
                "text": "User features",
                "type_str": "table",
                "content": "<table/>"
            },
            "TABREF3": {
                "html": null,
                "num": null,
                "text": "Distribution of classes across the topics.",
                "type_str": "table",
                "content": "<table><tr><td>Number of non-educational (NE), marginally educa-</td></tr><tr><td>tional (ME), educational (E) and strongly educational</td></tr><tr><td>(SE) materials.</td></tr></table>"
            }
        }
    }
}