File size: 58,616 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 |
{
"paper_id": "2021",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T01:06:23.697503Z"
},
"title": "BLAR: Biomedical Local Acronym Resolver",
"authors": [
{
"first": "William",
"middle": [],
"last": "Hogan",
"suffix": "",
"affiliation": {},
"email": "whogan@ucsd.edu"
},
{
"first": "Yoshiki",
"middle": [],
"last": "Vazquez Baeza",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of California",
"location": {
"addrLine": "San Diego",
"postCode": "92093",
"settlement": "La Jolla",
"region": "CA"
}
},
"email": ""
},
{
"first": "Yannis",
"middle": [],
"last": "Katsis",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "IBM Research-Almaden",
"location": {
"addrLine": "650 Harry Road",
"postCode": "95120",
"settlement": "San Jose",
"region": "CA"
}
},
"email": ""
},
{
"first": "Tyler",
"middle": [],
"last": "Baldwin",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "IBM Research-Almaden",
"location": {
"addrLine": "650 Harry Road",
"postCode": "95120",
"settlement": "San Jose",
"region": "CA"
}
},
"email": ""
},
{
"first": "Ho-Cheol",
"middle": [],
"last": "Kim",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "IBM Research-Almaden",
"location": {
"addrLine": "650 Harry Road",
"postCode": "95120",
"settlement": "San Jose",
"region": "CA"
}
},
"email": ""
},
{
"first": "Chun-Nan",
"middle": [],
"last": "Hsu",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of California",
"location": {
"addrLine": "San Diego",
"postCode": "92093",
"settlement": "La Jolla",
"region": "CA"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "NLP has emerged as an essential tool to extract knowledge from the exponentially increasing volumes of biomedical texts. Many NLP tasks, such as named entity recognition and named entity normalization, are especially challenging in the biomedical domain partly because of the prolific use of acronyms. Long names for diseases, bacteria, and chemicals are often replaced by acronyms. We propose Biomedical Local Acronym Resolver (BLAR), a highperforming acronym resolver that leverages state-of-the-art (SOTA) pre-trained language models to accurately resolve local acronyms in biomedical texts. We test BLAR on the Ab3P corpus and achieve state-of-the-art results compared to the current best-performing local acronym resolution algorithms and models.",
"pdf_parse": {
"paper_id": "2021",
"_pdf_hash": "",
"abstract": [
{
"text": "NLP has emerged as an essential tool to extract knowledge from the exponentially increasing volumes of biomedical texts. Many NLP tasks, such as named entity recognition and named entity normalization, are especially challenging in the biomedical domain partly because of the prolific use of acronyms. Long names for diseases, bacteria, and chemicals are often replaced by acronyms. We propose Biomedical Local Acronym Resolver (BLAR), a highperforming acronym resolver that leverages state-of-the-art (SOTA) pre-trained language models to accurately resolve local acronyms in biomedical texts. We test BLAR on the Ab3P corpus and achieve state-of-the-art results compared to the current best-performing local acronym resolution algorithms and models.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In the past decade, natural language processing (NLP) has greatly advanced in the biomedical domain. Given the troves of biomedical texts, NLP has emerged as a critical tool for knowledge extraction. NLP has been used to automatically analyze clinical notes, electronic medical records, biological literature, and other biomedical texts in the hopes of unearthing new knowledge and deeper insights.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Acronyms are especially common in science and even more so in biomedical publications, as authors regularly seek to shorten the long names for diseases, bacteria, and chemicals. Barnett and Doubleday (2020) documented acronym use in more than 24 million scientific article titles and 18 million scientific articles published between 1950 and 2019. They report that 19% of titles and 73% of abstracts contain acronyms. Of the more than one million unique acronyms in their data, 0.2% appeared regularly and most acronyms, 79%, appeared less than 10 times.",
"cite_spans": [
{
"start": 178,
"end": 206,
"text": "Barnett and Doubleday (2020)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Acronym resolution (AR) can be performed by either leveraging acronym definitions found in the text (referred to as local AR) or by consulting external resources, such as ontologies (known as disambiguation or global AR). While a lot of progress has been recently done on the latter, local AR has seen surprisingly little recent work. In particular, the SOTA approaches in local AR are rulebased or simple machine learning approaches from more than a decade ago. As a result, this task has not benefited from recent advances in transformers (Vaswani et al., 2017) . To address this issue, in this work we focus on local AR where we try to answer the question: Can transformers be leveraged to further improve traditional local AR approaches?",
"cite_spans": [
{
"start": 541,
"end": 563,
"text": "(Vaswani et al., 2017)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "To answer this question, we present Biomedical Local Acronym Resolver (BLAR); a transformerbased model designed to resolve local acronyms in biomedical texts. In particular, this work makes the following contributions:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "1. Design of a novel transformer-based model for local acronym resolution, which resolves acronyms through a combination of a two-step architecture and appropriate leveraging of pretrained language models. To the best of our knowledge, this is the first transformer-based approach for local AR.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "2. Experimental evaluation of BLAR against SOTA local AR approaches, showing that it outperforms the latter. In particular, evaluated on the Ab3P corpus (Sohn et al., 2008) , BLAR reaches an F1 score of 0.966 compared to 0.899 of the best performing existing approach.",
"cite_spans": [
{
"start": 153,
"end": 172,
"text": "(Sohn et al., 2008)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "There are a few challenges inherent in acronym resolution that make a simple dictionary-lookup and Our approach is applicable to both \"Local AR\" and \"Acronym Dictionary Compilation.\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background and Related Work",
"sec_num": "2"
},
{
"text": "other rule-based models less effective. First, shortform acronym representations are rarely unique. For instance, \"CD\" is an acronym for \"Crohn's disease\" and \"Cowden Disease.\" A simple dictionary lookup of \"CD\" using an acronym disease dictionary will produce ambiguous results and requires additional steps of acronym disambiguation. Moreover, the number of letters in a short-form may not match the number of words in the corresponding long-form (e.g. the short-form of \"systemic sclerosis\" is \"SSc\" ). Lastly, long-form entities can have complicated short-forms. For example, the short-form of \"heparin-induced thrombocytopenia type II\" is \"HIT type II,\" a short-form that shortens the first three words of the long-form and leaves the last two words unmodified. To address these challenges, approaches to acronym resolution have been developed and can be classified into three broad categories: local acronym resolution (Schwartz and Hearst, 2003; Sohn et al., 2008) , disambiguation acronym resolution (also referred to as non-local or global acronym resolution) (Jin et al., 2019; Jacobs et al., 2020) , and acronym dictionary compilation (Grossman et al., 2018). We refer to approaches that resolve acronyms by leveraging their definitions found in the containing text as local acronym resolution techniques. In contrast, non-local or global techniques resolve acronyms by using external resources. These typically target acronyms whose long-form is not contained within the text, which is common among more established acronyms, such as \"mRNA\" and \"DNA.\" Finally, acronym dic-tionary compilation refers to the creation of an acronym dictionary based on the source text or external ontologies, or a combination of the two. These three sub-categories of AR approaches are depicted in Figure 1 .",
"cite_spans": [
{
"start": 925,
"end": 952,
"text": "(Schwartz and Hearst, 2003;",
"ref_id": "BIBREF8"
},
{
"start": 953,
"end": 971,
"text": "Sohn et al., 2008)",
"ref_id": "BIBREF9"
},
{
"start": 1069,
"end": 1087,
"text": "(Jin et al., 2019;",
"ref_id": "BIBREF5"
},
{
"start": 1088,
"end": 1108,
"text": "Jacobs et al., 2020)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [
{
"start": 1791,
"end": 1799,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Background and Related Work",
"sec_num": "2"
},
{
"text": "Our approach specifically targets local acronym resolution and acronym dictionary compilation. Local acronyms appear as a pair of entities featuring a short-form (SF) entity and a corresponding longform (LF) entity. Historically, local acronym resolution has been handled by rule-based algorithms. From 2003 , Schwartz et al. (2003 and Sohn et al. (2008) demonstrated the best performance of local acronym resolution. They used a combination of hand-crafted filters to identify SF-LF pairs. Kuo et al. (2009) introduced the first local acronym resolution model that leveraged machine learning. It produced SOTA results with the help of four sets of hand-crafted features, including rule-based text filters. Yeganova et al. (2011) further improved upon local acronym resolution by introducing a hybrid machine learning and rulebase model that does not rely on labeled data. They extract potential SF-LF pairs from PubMed articles using rules similar to the rules developed by Sohn et al. and train a classifier to identify SF-LF pairs.",
"cite_spans": [
{
"start": 298,
"end": 307,
"text": "From 2003",
"ref_id": null
},
{
"start": 308,
"end": 331,
"text": ", Schwartz et al. (2003",
"ref_id": "BIBREF8"
},
{
"start": 336,
"end": 354,
"text": "Sohn et al. (2008)",
"ref_id": "BIBREF9"
},
{
"start": 491,
"end": 508,
"text": "Kuo et al. (2009)",
"ref_id": "BIBREF6"
},
{
"start": 707,
"end": 729,
"text": "Yeganova et al. (2011)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Background and Related Work",
"sec_num": "2"
},
{
"text": "Our approach to local acronym resolution is simple in its architecture yet novel in its application. Our two-stage model leverages transfer learning from modern, SOTA pretrained transformers and is able to learn the features of short-form and longform acronym pairs without the help of a predefined dictionary, hand-crafted features, filters, or rules. Our model processes batches of documents, such as abstracts from PubMed, and creates an acronym dictionary specific to each inputted document.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background and Related Work",
"sec_num": "2"
},
{
"text": "The intuition behind local acronym resolution is that authors of scientific publications commonly define the acronyms that they employ later on in the document. This is typically done by defining acronyms within the text in the form of pairs of short-form (SF) and corresponding long-form (LF) entities. We can then use the identified SF-LF acronym pairs to either resolve the acronyms appearing in the input document or populate an SF-LF dictionary that can be used to accurately resolve future uses of the SF versions of the acronyms in the remainder of the text. Step 2 showing the various tagged entities of a short and long-form acronym pair. We use a BILOU (Beginning, Inside, Last, Outside, Unit) tagging scheme (Ratinov and Roth, 2009) to identify long-form (LF) entities, short-form (SF) entities, and parenthesis (PR) enclosing a paired SF or LF entity.",
"cite_spans": [
{
"start": 719,
"end": 743,
"text": "(Ratinov and Roth, 2009)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Identifying the definitions of SF-LF pairs poses two major challenges: First, one has to identify the location in the text where the definition of an SF-LF pair is provided. Second, one has to identify the exact span (i.e., text) of both the short and long-form within the definition.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Two-step AR: Following the above structure, BLAR splits the problem into two separate subtasks:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "\u2022",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Step 1: Sentence Classification. Given the input text, identify sentences containing definitions of SF-LF pairs. This is modeled as a binary classification task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "\u2022",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Step 2: SF-LF Acronym Tagging: Given a sentence predicted to contain a definition of an SF-LF pair, identify the exact form (i.e., text) of the SF and LF entities. This is modeled as a token classification task, where each token in the sentence is classified as being part of an acronym short-form, acronym long-form, or the parenthesis enclosing a paired entity. Token classification follows the BILOU (Beginning, Inside, Last, Outside, and Unit) encoding scheme (Ratinov and Roth, 2009) , as shown in Figure 2 through a simple example.",
"cite_spans": [
{
"start": 464,
"end": 488,
"text": "(Ratinov and Roth, 2009)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [
{
"start": 503,
"end": 511,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Model architecture: The sentence classification model (Step 1) leverages transfer learning by fine-tuning the pretrained SciBERT model (Beltagy et al., 2019) for the specific task of sentence classification. The sentences that have been predicted as containing SF-LF pairs are given as input to the SF and LF tagging model (Step 2). The tagging model also leverages SciBERT by fine-tuning it on the SF and LF tagging task. To avoid exposure bias resulting from training on a set of perfect inputs (e.g. sentences containing acronym pairs as labeled in the dataset), we use the output from the sentence classification model from Step 1 to train the tagging model in Step 2. The output of the tagging model is a dictionary that can then be used to replace all the short-form acronyms with their corresponding long-forms within a single source text.",
"cite_spans": [
{
"start": 135,
"end": 157,
"text": "(Beltagy et al., 2019)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Model training: We developed BLAR using the BioADI corpus (Kuo et al., 2009) and tested it on the Ab3P corpus (Sohn et al., 2008) . BioADI includes 1,668 true SF-LF pairs from 1,200 annotated PubMed abstracts and Ab3P includes 1,221 true SF-LF pairs from 1,250 annotated PubMed abstracts. Both provide span-level data identifying short and long-form acronym pairs within PubMed abstracts and differ only in the articles selected for annotation. During development, we fine-tuned both our sentence and acronym token classifiers on the BioADI corpus randomly split into three subsets for training (80% of the corpus), validation (10% of the corpus), and testing (10% of the corpus). We use BioADI as a training dataset and Ab3P as a testing dataset to best compare our model's performance to existing SOTA benchmarks for local acronym resolution which use the same train/test splits. The BioADI and Ab3P corpora are described in Section 4. Since the models in both steps are fine-tuned versions of SciBERT, they are able to train fairly quick on CPUs.",
"cite_spans": [
{
"start": 58,
"end": 76,
"text": "(Kuo et al., 2009)",
"ref_id": "BIBREF6"
},
{
"start": 110,
"end": 129,
"text": "(Sohn et al., 2008)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Step 1 and Step 2 converged within eight epochs, taking roughly 10 hours and 2 hours to complete, respectively, on two Intel Xeon CPUs (E5-2640 v3 @ 2.60GH) with 16GB of RAM.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "Ablation study: To determine the importance of the 2-step architecture, we conduct an ablation study where we train a model to resolve acronyms without the help of a sentence classification step. This model is identical to the tagging model used in Step 2, only, it is trained on raw sentences that may or may not contain an acronym pair. This single-step architecture must simultaneously learn to detect and resolve an acronym pair. We refer to this model variation as \"BLAR (single step).\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Method",
"sec_num": "3"
},
{
"text": "BioADI: We use the BioADI (Kuo et al., 2009) corpus to train BLAR. It includes 1,668 true SF-LF pairs from 1,200 annotated PubMed abstracts.",
"cite_spans": [
{
"start": 26,
"end": 44,
"text": "(Kuo et al., 2009)",
"ref_id": "BIBREF6"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Datasets",
"sec_num": "4"
},
{
"text": "Ab3P: We use the Ab3P (Sohn et al., 2008) corpus for testing. It includes 1,221 true SF-LF pairs from 1,250 annotated PubMed abstracts.",
"cite_spans": [
{
"start": 22,
"end": 41,
"text": "(Sohn et al., 2008)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Datasets",
"sec_num": "4"
},
{
"text": "At the time of writing, both datasets are available for download on the BioC (Comeau et al., 2013) website.",
"cite_spans": [
{
"start": 77,
"end": 98,
"text": "(Comeau et al., 2013)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Datasets",
"sec_num": "4"
},
{
"text": "To measure BLAR's performance, we first compare it against SOTA local AR approaches. As explained in the Background and Previous Work section, to the best of our knowledge, local acronym resolution has not seen significant advances since 2009. More recent acronym resolution works have focused instead on disambiguation acronym resolution, still relying on simpler rule-based algorithms for local acronym resolution (Jin et al., 2019; Jacobs et al., 2020) . As a result, we compare BLAR to Kuo et al. (2009) , Sohn et al. (2008) , and Schwartz and Hearst (2003) , which represent the SOTA in local acronym resolution. Table 1 depicts the performance of BLAR against SOTA AR models. In this experiment, all models were trained on the BioADI dataset and tested on the Ab3P dataset. For each model, we evaluate Precision, Recall, and F1 score based on exact matches of long-form and short-form pairs. The results show that BLAR significantly outperforms all previous approaches, achieving an F1 score of 0.966 compared to 0.899 of the next best approach. We observe that, without a sentence classification step, the single-step BLAR model under-performs compared to the two-step architecture, highlighting the benefit of the sentence classification step in the full two-step architecture. Schwartz et al. (2003) 0.950 0.788 0.861 Sohn et al. (2008) 0.970 0.836 0.898 Kuo et al. (2009) 0.959 0.846 0.899 Yeganova et al. (2011) Model Output Analysis: Finally, to further understand the performance of BLAR, we perform an instance-level analysis of its output.",
"cite_spans": [
{
"start": 416,
"end": 434,
"text": "(Jin et al., 2019;",
"ref_id": "BIBREF5"
},
{
"start": 435,
"end": 455,
"text": "Jacobs et al., 2020)",
"ref_id": "BIBREF4"
},
{
"start": 490,
"end": 507,
"text": "Kuo et al. (2009)",
"ref_id": "BIBREF6"
},
{
"start": 510,
"end": 528,
"text": "Sohn et al. (2008)",
"ref_id": "BIBREF9"
},
{
"start": 535,
"end": 561,
"text": "Schwartz and Hearst (2003)",
"ref_id": "BIBREF8"
},
{
"start": 1286,
"end": 1308,
"text": "Schwartz et al. (2003)",
"ref_id": "BIBREF8"
},
{
"start": 1327,
"end": 1345,
"text": "Sohn et al. (2008)",
"ref_id": "BIBREF9"
},
{
"start": 1364,
"end": 1381,
"text": "Kuo et al. (2009)",
"ref_id": "BIBREF6"
},
{
"start": 1400,
"end": 1422,
"text": "Yeganova et al. (2011)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [
{
"start": 618,
"end": 625,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Results and Discussion",
"sec_num": "5"
},
{
"text": "Analyzing the correct predictions, we see that the model successfully overcomes some of the complex challenges inherent in acronym resolution. For example, it correctly resolves the acronyms \"SSc\" to \"systemic sclerosis\" and \"IUAG\" to \"intrauterine growth retardation.\" These examples show that BLAR learns to resolve short-forms that contain a different number of letters compared to the number of words in the corresponding long-form. In another example, BLAR correctly resolves \"HIT type II\" to \"heparin-induced thrombocytopenia type II\" which illustrates that the model was able to learn more complex acronyms that consist of a mix of short-form entities and complete words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "AR Model P R F1",
"sec_num": null
},
{
"text": "Moving to the incorrect predictions, we classify BLAR's errors into three categories: missed acronyms (false negatives), added acronyms (false positives), and modified acronyms (i.e., acronyms where the model correctly identifies a short-form but either truncates or extends the corresponding long-form).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "AR Model P R F1",
"sec_num": null
},
{
"text": "A majority of the errors come from modified acronyms. Analyzing the modified acronyms, we find that 63.7% of cases are long-forms expanded or truncated by a single word/token. We identify that many of the erroneously expanded long-forms add a word or words preceding the ground truth long-form. For example, in the text \". . . heat stroke by reducing iNOS-dependent nitric oxide (NO). . . \", BLAR identified \"iNOS-dependent nitric oxide\" as the long-form expansion of the short-form \"NO.\", instead of the correct \"nitric oxide.\"",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "AR Model P R F1",
"sec_num": null
},
{
"text": "Another common error within the modified acronyms category is a truncated long-form. For example, BLAR predicts the long-form of \"FVC\" to be \"forced vital capacity\" but the ground truth is \"forced expiratory volume in 1 s vital capacity.\" Here, BLAR predicts a simple long-form when the ground truth long-form is actually more complex. We plan to explore these insights in future work to further improve the model.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "AR Model P R F1",
"sec_num": null
},
{
"text": "Local acronym resolution has seen limited progress in recent years and has not benefited from the recent advancements in machine learning approaches. To address this problem, we develop BLAR; a deeplearning model that leverages a two-step architecture on top of pre-trained language models to identify SF-LF pairs in input documents. Our experimental results show that BLAR outperforms other local acronym resolution approaches and achieves state-of-the-art performance. We release BLAR and its source code for public use. As part of our future work, we will be exploring two threads: first, further improving the model based on our error analysis, and second, exploring how BLAR (which in this case has been fine-tuned for the scientific domain) can be extended to cover acronyms found in other domains. We believe future work could also focus on a hybrid model that leverages both deep-learning and rule-based algorithms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion and Future Work",
"sec_num": "6"
}
],
"back_matter": [
{
"text": "Thank you to the anonymous reviewers for their thoughtful comments and corrections. This work is supported by IBM Research AI through the AI Horizons Network.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgement",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Metaresearch: The growth of acronyms in the scientific literature. eLife",
"authors": [
{
"first": "Adrian",
"middle": [],
"last": "Barnett",
"suffix": ""
},
{
"first": "Zoe",
"middle": [],
"last": "Doubleday",
"suffix": ""
}
],
"year": 2020,
"venue": "",
"volume": "9",
"issue": "",
"pages": "",
"other_ids": {
"DOI": [
"10.7554/eLife.60080"
]
},
"num": null,
"urls": [],
"raw_text": "Adrian Barnett and Zoe Doubleday. 2020. Meta- research: The growth of acronyms in the scientific literature. eLife, 9:e60080.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "SciB-ERT: A pretrained language model for scientific text",
"authors": [
{
"first": "Iz",
"middle": [],
"last": "Beltagy",
"suffix": ""
},
{
"first": "Kyle",
"middle": [],
"last": "Lo",
"suffix": ""
},
{
"first": "Arman",
"middle": [],
"last": "Cohan",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP)",
"volume": "",
"issue": "",
"pages": "3615--3620",
"other_ids": {
"DOI": [
"10.18653/v1/D19-1371"
]
},
"num": null,
"urls": [],
"raw_text": "Iz Beltagy, Kyle Lo, and Arman Cohan. 2019. SciB- ERT: A pretrained language model for scientific text. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Lan- guage Processing (EMNLP-IJCNLP), pages 3615- 3620, Hong Kong, China. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Bioc: a minimalist approach to interoperability for biomedical text processing",
"authors": [
{
"first": "C",
"middle": [],
"last": "Donald",
"suffix": ""
},
{
"first": "Rezarta",
"middle": [],
"last": "Comeau",
"suffix": ""
},
{
"first": "Paolo",
"middle": [],
"last": "Islamaj Dogan",
"suffix": ""
},
{
"first": "Kevin",
"middle": [
"Bretonnel"
],
"last": "Ciccarese",
"suffix": ""
},
{
"first": "Martin",
"middle": [],
"last": "Cohen",
"suffix": ""
},
{
"first": "Florian",
"middle": [],
"last": "Krallinger",
"suffix": ""
},
{
"first": "Zhiyong",
"middle": [],
"last": "Leitner",
"suffix": ""
},
{
"first": "Yifan",
"middle": [],
"last": "Lu",
"suffix": ""
},
{
"first": "Fabio",
"middle": [],
"last": "Peng",
"suffix": ""
},
{
"first": "Manabu",
"middle": [],
"last": "Rinaldi",
"suffix": ""
},
{
"first": "Alfonso",
"middle": [],
"last": "Torii",
"suffix": ""
},
{
"first": "Karin",
"middle": [],
"last": "Valencia",
"suffix": ""
},
{
"first": "Thomas",
"middle": [
"C"
],
"last": "Verspoor",
"suffix": ""
},
{
"first": "Cathy",
"middle": [
"H"
],
"last": "Wiegers",
"suffix": ""
},
{
"first": "W. John",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Wilbur",
"suffix": ""
}
],
"year": 2013,
"venue": "Database : the journal of biological databases and curation",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"DOI": [
"10.1093/database/bat064"
]
},
"num": null,
"urls": [],
"raw_text": "Donald C. Comeau, Rezarta Islamaj Dogan, Paolo Ci- ccarese, Kevin Bretonnel Cohen, Martin Krallinger, Florian Leitner, Zhiyong Lu, Yifan Peng, Fabio Ri- naldi, Manabu Torii, Alfonso Valencia, Karin Ver- spoor, Thomas C. Wiegers, Cathy H. Wu, and W. John Wilbur. 2013. Bioc: a minimalist approach to interoperability for biomedical text processing. Database : the journal of biological databases and curation, 2013:bat064-bat064. 24048470[pmid].",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "A method for harmonization of clinical abbreviation and acronym sense inventories",
"authors": [
{
"first": "Lisa",
"middle": [
"V"
],
"last": "Grossman",
"suffix": ""
},
{
"first": "Elliot",
"middle": [
"G"
],
"last": "Mitchell",
"suffix": ""
},
{
"first": "George",
"middle": [],
"last": "Hripcsak",
"suffix": ""
},
{
"first": "Chunhua",
"middle": [],
"last": "Weng",
"suffix": ""
},
{
"first": "David",
"middle": [
"K"
],
"last": "Vawdrey",
"suffix": ""
}
],
"year": 2018,
"venue": "Journal of Biomedical Informatics",
"volume": "88",
"issue": "",
"pages": "62--69",
"other_ids": {
"DOI": [
"10.1016/j.jbi.2018.11.004"
]
},
"num": null,
"urls": [],
"raw_text": "Lisa V. Grossman, Elliot G. Mitchell, George Hripc- sak, Chunhua Weng, and David K. Vawdrey. 2018. A method for harmonization of clinical abbreviation and acronym sense inventories. Journal of Biomedi- cal Informatics, 88:62 -69.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Acronyms: identification, expansion and disambiguation",
"authors": [
{
"first": "Kayla",
"middle": [],
"last": "Jacobs",
"suffix": ""
},
{
"first": "Alon",
"middle": [],
"last": "Itai",
"suffix": ""
},
{
"first": "Shuly",
"middle": [],
"last": "Wintner",
"suffix": ""
}
],
"year": 2020,
"venue": "Annals of Mathematics and Artificial Intelligence",
"volume": "88",
"issue": "5",
"pages": "517--532",
"other_ids": {
"DOI": [
"10.1007/s10472-018-9608-8"
]
},
"num": null,
"urls": [],
"raw_text": "Kayla Jacobs, Alon Itai, and Shuly Wintner. 2020. Acronyms: identification, expansion and disam- biguation. Annals of Mathematics and Artificial In- telligence, 88(5):517-532.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Deep contextualized biomedical abbreviation expansion",
"authors": [
{
"first": "Qiao",
"middle": [],
"last": "Jin",
"suffix": ""
},
{
"first": "Jinling",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "Xinghua",
"middle": [],
"last": "Lu",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 18th BioNLP Workshop and Shared Task",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {
"DOI": [
"10.18653/v1/w19-5010"
]
},
"num": null,
"urls": [],
"raw_text": "Qiao Jin, Jinling Liu, and Xinghua Lu. 2019. Deep con- textualized biomedical abbreviation expansion. Pro- ceedings of the 18th BioNLP Workshop and Shared Task.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Bioadi: a machine learning approach to identifying abbreviations and definitions in biological literature",
"authors": [
{
"first": "Cheng-Ju",
"middle": [],
"last": "Kuo",
"suffix": ""
},
{
"first": "Maurice",
"middle": [
"Ht"
],
"last": "Ling",
"suffix": ""
},
{
"first": "Kuan-Ting",
"middle": [],
"last": "Lin",
"suffix": ""
},
{
"first": "Chun-Nan",
"middle": [],
"last": "Hsu",
"suffix": ""
}
],
"year": 2009,
"venue": "BMC Bioinformatics",
"volume": "10",
"issue": "15",
"pages": "",
"other_ids": {
"DOI": [
"10.1186/1471-2105-10-S15-S7"
]
},
"num": null,
"urls": [],
"raw_text": "Cheng-Ju Kuo, Maurice HT Ling, Kuan-Ting Lin, and Chun-Nan Hsu. 2009. Bioadi: a machine learn- ing approach to identifying abbreviations and defi- nitions in biological literature. BMC Bioinformatics, 10(15):S7.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Design challenges and misconceptions in named entity recognition",
"authors": [
{
"first": "Lev",
"middle": [],
"last": "Ratinov",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Roth",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Thirteenth Conference on Computational Natural Language Learning (CoNLL-2009)",
"volume": "",
"issue": "",
"pages": "147--155",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lev Ratinov and Dan Roth. 2009. Design chal- lenges and misconceptions in named entity recog- nition. In Proceedings of the Thirteenth Confer- ence on Computational Natural Language Learning (CoNLL-2009), pages 147-155, Boulder, Colorado. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "A simple algorithm for identifying abbreviation definitions in biomedical text",
"authors": [
{
"first": "Ariel",
"middle": [
"S"
],
"last": "Schwartz",
"suffix": ""
},
{
"first": "Marti",
"middle": [
"A"
],
"last": "Hearst",
"suffix": ""
}
],
"year": 2003,
"venue": "Pacific Symposium on Biocomputing. Pacific Symposium on Biocomputing",
"volume": "",
"issue": "",
"pages": "451--62",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ariel S. Schwartz and Marti A. Hearst. 2003. A sim- ple algorithm for identifying abbreviation definitions in biomedical text. Pacific Symposium on Biocom- puting. Pacific Symposium on Biocomputing, pages 451-62.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Abbreviation definition identification based on automatic precision estimates",
"authors": [
{
"first": "Sunghwan",
"middle": [],
"last": "Sohn",
"suffix": ""
},
{
"first": "Donald",
"middle": [
"C"
],
"last": "Comeau",
"suffix": ""
},
{
"first": "Won",
"middle": [],
"last": "Kim",
"suffix": ""
},
{
"first": "W. John",
"middle": [],
"last": "Wilbur",
"suffix": ""
}
],
"year": 2008,
"venue": "BMC bioinformatics",
"volume": "9",
"issue": "",
"pages": "402--402",
"other_ids": {
"DOI": [
"10.1186/1471-2105-9-402"
],
"PMCID": [
"PMC2576267"
]
},
"num": null,
"urls": [],
"raw_text": "Sunghwan Sohn, Donald C. Comeau, Won Kim, and W. John Wilbur. 2008. Abbreviation def- inition identification based on automatic preci- sion estimates. BMC bioinformatics, 9:402-402. PMC2576267[pmcid].",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Attention is all you need",
"authors": [
{
"first": "Ashish",
"middle": [],
"last": "Vaswani",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Shazeer",
"suffix": ""
},
{
"first": "Niki",
"middle": [],
"last": "Parmar",
"suffix": ""
},
{
"first": "Jakob",
"middle": [],
"last": "Uszkoreit",
"suffix": ""
},
{
"first": "Llion",
"middle": [],
"last": "Jones",
"suffix": ""
},
{
"first": "Aidan",
"middle": [
"N"
],
"last": "Gomez",
"suffix": ""
},
{
"first": "Illia",
"middle": [],
"last": "Kaiser",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Polosukhin",
"suffix": ""
}
],
"year": 2017,
"venue": "Advances in Neural Information Processing Systems",
"volume": "30",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, \u0141 ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in Neural Information Pro- cessing Systems, volume 30. Curran Associates, Inc.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Machine learning with naturally labeled data for identifying abbreviation definitions",
"authors": [
{
"first": "Lana",
"middle": [],
"last": "Yeganova",
"suffix": ""
},
{
"first": "Donald",
"middle": [],
"last": "Comeau",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Wilbur",
"suffix": ""
}
],
"year": 2011,
"venue": "BMC bioinformatics",
"volume": "12",
"issue": "3",
"pages": "",
"other_ids": {
"DOI": [
"10.1186/1471-2105-12-S3-S6"
]
},
"num": null,
"urls": [],
"raw_text": "Lana Yeganova, Donald Comeau, and W. Wilbur. 2011. Machine learning with naturally labeled data for identifying abbreviation definitions. BMC bioinfor- matics, 12 Suppl 3:S6.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"num": null,
"uris": null,
"text": "Sub-tasks of acronym resolution (AR).",
"type_str": "figure"
},
"FIGREF1": {
"num": null,
"uris": null,
"text": "Sample output of",
"type_str": "figure"
},
"TABREF1": {
"num": null,
"content": "<table><tr><td>: Evaluation results of BLAR against SOTA</td></tr><tr><td>local acronym resolution models. All models, save</td></tr><tr><td>Yeganova et al., were trained on BioADI and tested</td></tr><tr><td>on Ab3P. Yeganova et al. is trained on 1M automati-</td></tr><tr><td>cally extracted potential SF-LF pairs from PubMed ab-</td></tr><tr><td>stracts.</td></tr></table>",
"text": "",
"html": null,
"type_str": "table"
}
}
}
} |