Skip to content

Latest commit

 

History

History
126 lines (114 loc) · 4.99 KB

README.md

File metadata and controls

126 lines (114 loc) · 4.99 KB
Raw

CFS3

The Communicative-Function-labeled Semantic Scholar Sentence Dataset (CFS3) is a distantly-labeled dataset that includes 100,016 sentences with 77 communicative-function labels.
All sentences are derived from S2ORC.

If you find this useful, please cite our work.

citation

@inproceedings{sugimoto-aizawa-2022-incorporating,
    title = "Incorporating the Rhetoric of Scientific Language into Sentence Embeddings using Phrase-guided Distant Supervision and Metric Learning",
    author = "Sugimoto, Kaito  and
      Aizawa, Akiko",
    booktitle = "Proceedings of the Third Workshop on Scholarly Document Processing",
    month = oct,
    year = "2022",
    address = "Gyeongju, Republic of Korea",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2022.sdp-1.7",
    pages = "54--68",
    abstract = "Communicative functions are an important rhetorical feature of scientific writing. Sentence embeddings that contain such features are highly valuable for the argumentative analysis of scientific documents, with applications in document alignment, recommendation, and academic writing assistance. Moreover, embeddings can provide a possible solution to the open-set problem, where models need to generalize to new communicative functions unseen at training time. However, existing sentence representation models are not suited for detecting functional similarity since they only consider lexical or semantic similarities. To remedy this, we propose a combined approach of distant supervision and metric learning to make a representation model more aware of the functional part of a sentence. We first leverage an existing academic phrase database to label sentences automatically with their functions. Then, we train an embedding model to capture similarities and dissimilarities from a rhetorical perspective. The experimental results demonstrate that the embeddings obtained from our model are more advantageous than existing models when retrieving functionally similar sentences. We also provide an extensive analysis of the performance differences between five metric learning objectives, revealing that traditional methods (e.g., softmax cross-entropy loss and triplet loss) outperform state-of-the-art techniques.",
}

Each label number corresponds to a total of 80 different functional classes included in one of the links below (accessed Nov. 2021).

Label Distribution

Labels 15, 29, and 39 are assigned to no sentence.

label CFS3 CFS3_train CFS3_valid
01 5212 4170 1042
02 520 416 104
03 334 267 67
04 5165 4132 1033
05 726 581 145
06 283 226 57
07 140 112 28
08 541 433 108
09 981 785 196
10 4101 3281 820
11 1735 1388 347
12 1339 1071 268
13 300 240 60
14 1102 882 220
15 0 0 0
16 142 114 28
17 493 394 99
18 73 58 15
19 134 107 27
20 172 138 34
21 211 169 42
22 10686 8549 2137
23 10 8 2
24 59 47 12
25 184 147 37
26 107 86 21
27 4 3 1
28 68 54 14
29 0 0 0
30 245 196 49
31 2 2 0
32 526 421 105
33 1694 1355 339
34 4394 3515 879
35 340 272 68
36 2950 2360 590
37 1774 1419 355
38 823 658 165
39 0 0 0
40 426 341 85
41 257 206 51
42 69 55 14
43 71 57 14
44 160 128 32
45 1988 1590 398
46 49 39 10
47 699 559 140
48 11873 9498 2375
49 2404 1923 481
50 1033 826 207
51 1642 1314 328
52 9 7 2
53 203 162 41
54 139 111 28
55 66 53 13
56 12 10 2
57 42 34 8
58 6 5 1
59 91 73 18
60 416 333 83
61 776 621 155
62 114 91 23
63 3096 2477 619
64 543 434 109
65 5463 4370 1093
66 982 786 196
67 3800 3040 760
68 1254 1003 251
69 746 597 149
70 121 97 24
71 23 18 5
72 628 502 126
73 3243 2594 649
74 447 358 89
75 1460 1168 292
76 2805 2244 561
77 2091 1673 418
78 169 135 34
79 2466 1973 493
80 564 451 113

License

CC BY-NC 4.0