File size: 70,341 Bytes
6fa4bc9 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 | {
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T13:05:02.142878Z"
},
"title": "TermEval 2020: RACAI's automatic term extraction system",
"authors": [
{
"first": "Vasile",
"middle": [],
"last": "P\u0103i\u0219",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Romanian Academy CASA ACADEMIEI",
"location": {
"addrLine": "13 \"Calea 13 Septembrie\"",
"postCode": "050711",
"settlement": "Bucharest",
"country": "ROMANIA"
}
},
"email": "vasile@racai.ro"
},
{
"first": "Radu",
"middle": [],
"last": "Ion",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Romanian Academy CASA ACADEMIEI",
"location": {
"addrLine": "13 \"Calea 13 Septembrie\"",
"postCode": "050711",
"settlement": "Bucharest",
"country": "ROMANIA"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "This paper describes RACAI's automatic term extraction system, which participated in the TermEval 2020 shared task on English monolingual term extraction. We discuss the system architecture, some of the challenges that we faced as well as present our results in the English competition.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "This paper describes RACAI's automatic term extraction system, which participated in the TermEval 2020 shared task on English monolingual term extraction. We discuss the system architecture, some of the challenges that we faced as well as present our results in the English competition.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Automatic term extraction, also known as ATE, is a wellknown task within the domain of natural language processing. Given a text (this can be either a fragment or an entire corpus), an automatic term extractor system will produce a list of terms (single or multiword expressions) characteristic for the domain of text.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "Felber, in the \"Terminology Manual\" (Felber, 1984) , defines a term as \"any conventional symbol representing a concept defined in a subject field\". Nevertheless, considering current practice in natural language processing tasks, it is not always possible to give a general definition applicable for the workings of a term extractor. One question is whether or not to include named entities as part of the identified terms. This problem is also raised by the organizers of the TermEval 2020 shared task, each system being evaluated twice, once including and once excluding named entities 1 . Furthermore, since named entity recognizers can be trained on many classes (such as diseases or chemicals for example), another potential question is what kinds of entities (if any) can be included as part of the identified terms. However, an agreement must be made that all identified terms must be specific to the domain of the analyzed text, regardless of inclusion or not of named entities. For example, in the shared task's provided training dataset, the named entity \"United States Dressage Federation\" is included as a term in the \"equestrian\" section.",
"cite_spans": [
{
"start": 36,
"end": 50,
"text": "(Felber, 1984)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "The present paper presents our attempt at constructing an automatic term extraction system in the context of the TermEval 2020 shared task on monolingual term extraction (Rigouts Terryn et al., 2020) . We start by presenting related research, then continue with the description of our system and finally present concluding remarks.",
"cite_spans": [
{
"start": 170,
"end": 199,
"text": "(Rigouts Terryn et al., 2020)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1."
},
{
"text": "The usefulness of the term identification process is both in its own use, such as creation of document indices, and as a pre-processing step in other more advanced processes, such as machine translation. Furthermore, the output produced by an automatic system can be manually validated by a human user in order to remove irrelevant terms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "2."
},
{
"text": "Traditional approaches for ATE (Kageura, 1998 ) make use of statistical features such as word frequency or \"termhood\" (degree of relatedness of a proposed term to the domain) metrics. Additionally, information such as part of speech can be used to further filter candidate terms. Term formalization attempts can be identified in the literature as early as e.g. 1996, when Frantzi and Ananiadou (1996) defined C-value as a basic measure of termhood, a principle we have also used in one of our algorithms. In this section, we will briefly mention the inner workings of some existing term extraction algorithms that we used in our term extraction system. For a detailed coverage of this rather vast sub-domain of NLP, the reader is referred to e.g. Pazienza et al. (2005) or the more recent Firoozeh et al. (2019) .",
"cite_spans": [
{
"start": 31,
"end": 45,
"text": "(Kageura, 1998",
"ref_id": "BIBREF10"
},
{
"start": 372,
"end": 400,
"text": "Frantzi and Ananiadou (1996)",
"ref_id": "BIBREF7"
},
{
"start": 747,
"end": 769,
"text": "Pazienza et al. (2005)",
"ref_id": "BIBREF14"
},
{
"start": 789,
"end": 811,
"text": "Firoozeh et al. (2019)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "2."
},
{
"text": "TextRank (Mihalcea and Tarau, 2004 ) is a term extraction algorithm using a graph representation of the text in which each word is a node and an edge is created between words collocated within a certain window of words. Based on the number of links to each node a score is computed similar to the PageRank algorithm (Brin and Page, 1998) . Further filtering is performed based on the part of speech of the words. The graph is created based on single words. However, as the last step of the algorithm a reconstruction of multi-word terms is performed if multiple single word terms are collocated in the sentence. RAKE, an acronym for Rapid Automatic Keyword Extraction (Rose et al., 2010) , combines graph measures such as the degree (number of connected edges) with statistical measures such as word frequency. Furthermore, RAKE uses a strategy similar to TextRank for combining single words that occur together at least twice into a multi-word term. An interesting idea deriving from the RAKE paper is the importance of the stop words list used. In this context, it is mentioned that FOX (Fox, 1989) stop list produces an increase in the F1 score for the RAKE algorithm. An improvement over the initial RAKE algorithm is described in Gupta et al. (2016) . Campos et al. (2020) present YAKE, which makes use of statistical features. According to their analysis 2 it is comparable or even better in some cases to previous stateof-the-art methods. In the HAMLET system (Rigouts Terryn et al., 2019) a number of 152 features are computed on each candidate term and a binary decision tree classifier is trained. Candidates are determined based on their part of speech, but the patterns of occurrence are determined automatically based on training data.",
"cite_spans": [
{
"start": 9,
"end": 34,
"text": "(Mihalcea and Tarau, 2004",
"ref_id": "BIBREF12"
},
{
"start": 316,
"end": 337,
"text": "(Brin and Page, 1998)",
"ref_id": "BIBREF1"
},
{
"start": 668,
"end": 687,
"text": "(Rose et al., 2010)",
"ref_id": "BIBREF18"
},
{
"start": 1089,
"end": 1100,
"text": "(Fox, 1989)",
"ref_id": "BIBREF6"
},
{
"start": 1235,
"end": 1254,
"text": "Gupta et al. (2016)",
"ref_id": "BIBREF8"
},
{
"start": 1257,
"end": 1277,
"text": "Campos et al. (2020)",
"ref_id": "BIBREF2"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related work",
"sec_num": "2."
},
{
"text": "The dataset proposed for the TermEval task is described in detail in the task paper (Rigouts Terryn et al., 2020) . However, several aspects must be mentioned. It is comprised of 4 domains: wind energy ('wind'), corruption ('corp'), horse dressage ('equi') and heart failure ('hf'). The first 3 domains were provided with annotations for training purposes, while the heart failure domain was used for testing. All the domains were made available in English, French and Dutch.",
"cite_spans": [
{
"start": 84,
"end": 113,
"text": "(Rigouts Terryn et al., 2020)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "For the purposes of our experiments, we focused on the English version of the corpus. However, we tried to keep our algorithms independent of the actual language being used. Towards this end, we used only resources normally available for many languages, such as annotations and stop words, and did not create any rules or patterns specific to the English language.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "One of the primary processing operations was to annotate the corpus with part-of-speech and lemma information. For this purpose, we used Stanford CoreNLP (Manning et al., 2014) . Furthermore, we precomputed statistical indicators based on the corpus, such as n-gram frequency, document frequency and letters used (in some cases terms contained non-English letters). Statistics were computed for both the corpus and the provided training annotations.",
"cite_spans": [
{
"start": 154,
"end": 176,
"text": "(Manning et al., 2014)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "Unfortunately, the corpus is not balanced with respect to the different domains. Therefore, some statistical indicators may be less meaningful. For example, the corruption part of the corpus contains 12 annotated texts with an additional 12 texts provided without annotations. However, the equestrianism part contains 34 annotated text files and 55 unannotated documents. Furthermore, the evaluation section on heart failure contains 190 files. This seems to suggest that indicators like document frequency (the number of documents containing a certain word/expression) may be more meaningful for certain sections and less meaningful for others.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "More statistics regarding the English domains of the corpus are presented in Table 1 One of the characteristics specific only to the wind energy section of the corpus is the presence of mathematical formulas in some of the files. We could not identify an easy way to automatically remove them and did not want to manually perform this action. For example, \"CP\" is considered a term and it also appears in some formulas. Furthermore, there are lines of text presumably between formulas which look similar to a formula, like \"CP ,max CT CTr\" or full lines of text containing embedded formulas. Even more, the term \"PCO2\", indicated in the gold annotations, seems to only appear inside a formula (\"PCO2 = TCO2 -HCO2 PCO2\"). Therefore, in order to avoid removal of potentially useful portions of text, the files were used as they were provided.",
"cite_spans": [],
"ref_spans": [
{
"start": 77,
"end": 84,
"text": "Table 1",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "Given these discrepancies between the different domain sub-corpora, it was our assumption, from the beginning, that different algorithms will obtain different results on each of the domains. Therefore, we started first by analyzing the results provided by known algorithms on the training parts of the corpus. These results are presented in Tables 2, 3 , 4 and are compared against the provided annotations with named entities included. In these tables, the algorithm with the best F1 score in each section is marked in bold. The \"1W\" specification besides an algorithm denotes the score for single word terms.",
"cite_spans": [],
"ref_spans": [
{
"start": 341,
"end": 352,
"text": "Tables 2, 3",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "In accordance with our previous observation, because of the imbalances between the different sections of the corpus, from Table 2 it can easily be seen that most of the algorithms perform better on the \"equi\" section and worse on the other sections. In some cases, there are even extreme differences. For example, the YAKE implementation gives on multi-word expressions an F1 score of 22.3 on the \"equi\" section and only 5.94 on the \"wind\" section. This is improved for single word expressions with 12% on the \"equi\" section and less then 3% for the other sections. ",
"cite_spans": [],
"ref_spans": [
{
"start": 122,
"end": 129,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Dataset and basic processing",
"sec_num": "3."
},
{
"text": "Looking at the above tables, two observations can be made: a) no single system performs best on all three sections; b) systems tend to balance precision and recall, but in extreme cases they prefer either precision (for example the YAKE method in \"corp\" and \"wind\" sections) or recall (for example the RAKE method).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "A first idea that we explored was to implement a voting mechanism between the systems. However, the results presented only slight improvements. Without a complete and in-depth analysis, we concluded that each system was good at identifying certain terms (based on their pattern of occurrence) but performing badly for other terms. Therefore, we decided to extend the basic system and implement additional algorithms that would try to complement and extend the previous ones, by using new methods and finally use the same voting mechanism.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "The first algorithm, PLEARN (from \"pattern learn\") is trying to identify patterns based on statistics computed on the train set annotations and their appearance in context. We used the following features: letters accepted in annotations (for example there is no term using \",\"), stop words accepted at start or end of a term (for example there is no term starting or ending with \"and\"), stop words accepted inside multi word terms, stop words accepted before or after a term (for example \"and\" usually is not contained within a term but rather it separates two distinct terms, thus appearing before or after a term), suffixes of words other than stop words present in terms (usually we tend to find nouns as terms, but we tried not to impose this condition, thus we only checked the suffixes of words).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "For the purpose of the algorithm, all information was extracted automatically from the training set and no manual conditions or word lists were created. One immediate problem with the algorithm is that the training set did not provide the actual position of the term. Therefore, if the same word or multi-word expression was used both as term and as a non-term then the feature extraction part was not able to identify this case. Nevertheless, the algorithm was able to produce the good recall that we were expecting, presented in A second algorithm used a clustering approach, thus we'll refer to it as \"CLUS\" for the purposes of this paper. In this case we worked under the assumption that terms belonging to a particular domain will tend to cluster together because they will be related in meaning. In order to model this relation, we represented the words using word embeddings and used the cosine distance. For the clustering algorithm, we implemented a DBSCAN algorithm (Ester et al., 1996) .",
"cite_spans": [
{
"start": 976,
"end": 996,
"text": "(Ester et al., 1996)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "The input for the clustering algorithm was composed of the terms identified by the PLEARN algorithm. From these terms we kept only the single word terms. Furthermore, we decided to use an approach similar to the one used in TextRank to compose at the end multi-word terms based on the colocation of single word terms. This last operation was done in a post-processing step.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "For the word embedding representation we considered necessary to use a model trained on a large enough corpus to allow for words to be used in different domains, including those of interest for this work. Therefore, we decided to use a word embeddings model trained on the Open American National Corpus (Ide, 2008) . Furthermore, due to the relatively short time available for the task participation, we decided to use a pre-trained model 3 . Results are given in Table 6 .",
"cite_spans": [
{
"start": 303,
"end": 314,
"text": "(Ide, 2008)",
"ref_id": "BIBREF9"
}
],
"ref_spans": [
{
"start": 464,
"end": 471,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "This algorithm already has a much better F1 score for single word terms then all the other algorithms tested. In the case of the \"wind\" section the F1 score is almost double (45.02%) then the best previous result (22.79%). Table 6 : Precision, Recall, F1 measures for the CLUS algorithm on the training parts of the corpus",
"cite_spans": [],
"ref_spans": [
{
"start": 223,
"end": 230,
"text": "Table 6",
"ref_id": null
}
],
"eq_spans": [],
"section": "System Architecture",
"sec_num": "4."
},
{
"text": "Since the CLUS algorithm works on single word terms and only in the post-processing step combines them to create multi-word terms, we decided to work on a third algorithm that would work directly with multi-word expression candidates.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "The third (and last) algorithm that we developed is called WEMBF (word embeddings filtered) and, as its name implies, uses the word embeddings vector representation of words to measure the termhood of each word. The algorithm executes the following steps: 1) Tokenizes and POS tags all text files of the specified domain of the corpus, using the NLTK Python library (Bird et al., 2009) ;",
"cite_spans": [
{
"start": 366,
"end": 385,
"text": "(Bird et al., 2009)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "2) Extracts all NPs from the domain sub-corpus, using simple prenominal-nominal patterns, including all prepositional phrases headed by the preposition 'of', which are almost always attached to the previous NP. Furthermore, it deletes any determiners that start NPs and removes URLs, emails, numbers and other entities considered to be irrelevant for the term extraction task;",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "3) For each content word (i.e. nouns, adjectives, adverbs and verbs) of each NP, computes a cosine distance between two word embeddings vectors. The first vector is obtained from training on a \"general\"-domain corpus containing news, literature, sports, etc., being careful not to include texts from the domain of interest. The second vector is obtained from training only on the domain of interest (e.g. 'wind'); 4) Score each NP by averaging the previously computed cosine distance of its member content words.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "Step 4 of the WEMBF algorithm gives us a preliminary term list on the assumption that the larger the cosine distance of the general and domain word embeddings vectors is, the more likely is that the word is a term in the domain of interest. However, the obtained list contains too many NPs which makes it perform poorly in terms of precision. Thus, we decided to remove some term NPs from this initial list, using the following filters: a) Only keep NPs which appear (are embedded) in other NPs from the preliminary term list (Frantzi and Ananiadou, 1996) . The number of occurrences (in other NPs) is kept for each surviving NP to be rescored later;",
"cite_spans": [
{
"start": 526,
"end": 555,
"text": "(Frantzi and Ananiadou, 1996)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "b) Remove all single-word terms that appear as head nouns in other NPs on the assumption that if they can be modified, they are too general to be kept as terms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "The termhood score of each NP in the final list is modified by multiplying the following indicators: the original score of the NP, the number of words in the NP, the number of NPs in which this NP appeared.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "Thus, if an NP has more words, it appeared in many other NPs and its average cosine distance (between the general domain and the domain of interest) of its member content words is higher, the NP is more likely to be a term.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "Results of the WEMBF term extraction algorithm are given in Table 8 Table 8 . Precision, Recall, F1 measures for the WEMBF algorithm on the training parts of the corpus",
"cite_spans": [],
"ref_spans": [
{
"start": 60,
"end": 67,
"text": "Table 8",
"ref_id": null
},
{
"start": 68,
"end": 75,
"text": "Table 8",
"ref_id": null
}
],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "The WEMBF algorithm has a performance similar to the PLEARN algorithm for single words, even though with a more balanced precision and recall, but better performance for multi-word terms.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "The final step in our approach was to construct an ensemble module that takes the annotations from different algorithms and combines them together via a voting scheme. This is presented schematically in Figure 1 . Each algorithm is fed into the voting module, having one vote for the final result. An exception is in the case of PLEARN and CLUS algorithms which are linked together and thus constitute a single vote.",
"cite_spans": [],
"ref_spans": [
{
"start": 203,
"end": 211,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "P%",
"sec_num": null
},
{
"text": "Once the test set annotations were released, we were able to evaluate our system, including all the other algorithms on the final data. When comparing this information with results based on the different training sections, we must keep in mind the peculiarities of each section of the corpus, as presented in Table 1 above. Evaluation results on the \"heart failure\" section are presented in Table 9 .",
"cite_spans": [],
"ref_spans": [
{
"start": 309,
"end": 316,
"text": "Table 1",
"ref_id": "TABREF1"
},
{
"start": 391,
"end": 398,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "System evaluation",
"sec_num": "5."
},
{
"text": "Our CLUS algorithm performed best on the single word terms giving an F1 score of 53.48 with balanced precision and recall. Furthermore, the PLEARN algorithm produced the best recall, which was to be expected since it was designed especially for this purpose. However, the final algorithm with the combination of all of them did perform better on the multi-word terms, this being reflected in the final F1 score. Table 9 . Precision, Recall, F1 measures of different algorithms on the evaluation set (\"heart failure\").",
"cite_spans": [],
"ref_spans": [
{
"start": 412,
"end": 419,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "System evaluation",
"sec_num": "5."
},
{
"text": "This paper presented our system proposal 4 for the TermEval 2020 shared task. We started by investigating the performance of existing algorithms. Then went on and created three new algorithms: PLEARN, CLUS and WEMBF as described in section 4. Finally, we constructed an ensemble module, based on voting, which combined the results of all the algorithms in order to produce the final results. Evaluation on the \"heart failure\" dataset is presented in Table 9 above.",
"cite_spans": [],
"ref_spans": [
{
"start": 450,
"end": 457,
"text": "Table 9",
"ref_id": null
}
],
"eq_spans": [],
"section": "Conclusions and future work",
"sec_num": "6."
},
{
"text": "The approach behind the ACTER dataset, of building a term annotated corpus in multiple languages is very interesting and it was extremely helpful for building our automatic term extractor system. It is our hope that this or a similar approach could be used for Romanian language as well. In this context, we envisage extending our term extractor to support Romanian language and further include it in the RELATE platform (P\u0103i\u0219 et al., 2019) dedicated to processing Romanian language.",
"cite_spans": [
{
"start": 421,
"end": 440,
"text": "(P\u0103i\u0219 et al., 2019)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and future work",
"sec_num": "6."
},
{
"text": "We managed to successfully use pre-trained word embeddings on a large corpus for our CLUS algorithm. This proves that transfer learning is a possibility that should be explored also in the field of term extraction. Therefore, amongst our future work we'll try to use the same approach for the Romanian language, by using pretrained word embeddings (P\u0103i\u0219 and Tufi\u0219, 2018) on the Reference Corpus of Contemporary Romanian Language (CoRoLa) (Mititelu et al., 2018) .",
"cite_spans": [
{
"start": 348,
"end": 370,
"text": "(P\u0103i\u0219 and Tufi\u0219, 2018)",
"ref_id": null
},
{
"start": 438,
"end": 461,
"text": "(Mititelu et al., 2018)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and future work",
"sec_num": "6."
},
{
"text": "Part of this work was conducted in the context of the ReTeRom project. Part of this work was conducted in the context of the Marcell project.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": "7."
},
{
"text": "https://termeval.ugent.be/task-evaluation/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/LIAAD/yake",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://data.world/jaredfern/oanc-word-embeddings",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "https://github.com/racai-ai/TermEval2020",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Natural Language Processing with Python ---Analyzing Text with the Natural Language Toolkit",
"authors": [
{
"first": "S",
"middle": [],
"last": "Bird",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Klein",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Loper",
"suffix": ""
}
],
"year": 2009,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bird, S., Klein, E. and Loper, E. (2009). Natural Language Processing with Python ---Analyzing Text with the Natural Language Toolkit. O'Reilly Media; available online at http://www.nltk.org/book_1ed/.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "The anatomy of a large-scale hypertextual Web search engine. Computer Networks and ISDN Systems",
"authors": [
{
"first": "S",
"middle": [],
"last": "Brin",
"suffix": ""
},
{
"first": "L",
"middle": [],
"last": "Page",
"suffix": ""
}
],
"year": 1998,
"venue": "",
"volume": "30",
"issue": "",
"pages": "1--7",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Brin, S. and Page, L. (1998). The anatomy of a large-scale hypertextual Web search engine. Computer Networks and ISDN Systems, 30(1-7).",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "YAKE! Keyword Extraction from Single Documents using Multiple Local Features",
"authors": [
{
"first": "R",
"middle": [],
"last": "Campos",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Mangaravite",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Pasquali",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Jatowt",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Jorge",
"suffix": ""
},
{
"first": "C",
"middle": [],
"last": "Nunes",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Jatowt",
"suffix": ""
}
],
"year": 2020,
"venue": "Information Sciences Journal. Elsevier",
"volume": "509",
"issue": "",
"pages": "257--289",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Campos, R., Mangaravite, V., Pasquali, A., Jatowt, A., Jorge, A., Nunes, C. and Jatowt, A. (2020). YAKE! Keyword Extraction from Single Documents using Multiple Local Features. In Information Sciences Journal. Elsevier, Vol 509, pp 257-289.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "A density-based algorithm for discovering clusters in large spatial databases with noise",
"authors": [
{
"first": "M",
"middle": [],
"last": "Ester",
"suffix": ""
},
{
"first": "H",
"middle": [
"P"
],
"last": "Kriegel",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Sander",
"suffix": ""
},
{
"first": "X",
"middle": [],
"last": "Xu",
"suffix": ""
}
],
"year": 1996,
"venue": "Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96)",
"volume": "",
"issue": "",
"pages": "226--231",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ester, M., Kriegel, H. P., Sander, J. and Xu, X. (1996). A density-based algorithm for discovering clusters in large spatial databases with noise. In Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96),pp 226-231.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Terminology Manual. Paris: International Information Centre for Terminology",
"authors": [
{
"first": "H",
"middle": [],
"last": "Felber",
"suffix": ""
}
],
"year": 1984,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Felber, H. (1984). Terminology Manual. Paris: International Information Centre for Terminology.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Keyword extraction: Issues and methods",
"authors": [
{
"first": "N",
"middle": [],
"last": "Firoozeh",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Nazarenko",
"suffix": ""
},
{
"first": "F",
"middle": [],
"last": "Alizon",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Daille",
"suffix": ""
}
],
"year": 2019,
"venue": "Natural Language Engineering",
"volume": "",
"issue": "",
"pages": "1--33",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Firoozeh, N., Nazarenko, A., Alizon, F. and Daille, B. (2019). Keyword extraction: Issues and methods. Natural Language Engineering, pages 1-33, Cambridge University Press.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "A stop list for general text",
"authors": [
{
"first": "C",
"middle": [],
"last": "Fox",
"suffix": ""
}
],
"year": 1989,
"venue": "ACM SIGIR Forum",
"volume": "24",
"issue": "",
"pages": "19--21",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Fox, C. (1989). A stop list for general text. ACM SIGIR Forum, vol. 24, pp. 19-21. ACM, New York, USA.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Extracting Nested Collocations",
"authors": [
{
"first": "K",
"middle": [
"T"
],
"last": "Frantzi",
"suffix": ""
},
{
"first": "Sophia",
"middle": [],
"last": "Ananiadou",
"suffix": ""
}
],
"year": 1996,
"venue": "Proceedings of the 16th conference on Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "41--46",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Frantzi, K. T. and Ananiadou, Sophia. (1996) Extracting Nested Collocations. In Proceedings of the 16th conference on Computational Linguistics -Volume 1, pages 41-46. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Rake-pmi automated keyphrase extraction: An unsupervised approach for automated extraction of keyphrases",
"authors": [
{
"first": "S",
"middle": [],
"last": "Gupta",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Mittal",
"suffix": ""
},
{
"first": "A",
"middle": [],
"last": "Kumar",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the International Conference on Informatics and Analytics",
"volume": "",
"issue": "",
"pages": "1--6",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Gupta, S., Mittal, N., & Kumar, A. (2016). Rake-pmi automated keyphrase extraction: An unsupervised approach for automated extraction of keyphrases. In Proceedings of the International Conference on Informatics and Analytics, pp. 1-6.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "The American National Corpus: Then, Now, and Tomorrow",
"authors": [
{
"first": "N",
"middle": [],
"last": "Ide",
"suffix": ""
}
],
"year": 2008,
"venue": "Selected Proceedings of the 2008 HCSNet Workshop on Designing the Australian National Corpus: Mustering Languages, Cascadilla Proceedings Project",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ide, N. (2008). The American National Corpus: Then, Now, and Tomorrow. In Michael Haugh, Kate Burridge, Jean Mulder and Pam Peters (eds.), Selected Proceedings of the 2008 HCSNet Workshop on Designing the Australian National Corpus: Mustering Languages, Cascadilla Proceedings Project, Sommerville, MA.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Methods of automatic term recognition",
"authors": [
{
"first": "K",
"middle": [],
"last": "Kageura",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Umino",
"suffix": ""
}
],
"year": 1998,
"venue": "Terminology",
"volume": "3",
"issue": "2",
"pages": "259--289",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kageura, K.; Umino, B. (1998). Methods of automatic term recognition. Terminology. 3(2):259-289.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "The Stanford CoreNLP Natural Language Processing Toolkit",
"authors": [
{
"first": "C",
"middle": [
"D"
],
"last": "Manning",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Surdeanu",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Bauer",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Finkel",
"suffix": ""
},
{
"first": "S",
"middle": [
"J"
],
"last": "Bethard",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Mcclosky",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations",
"volume": "",
"issue": "",
"pages": "55--60",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Manning, C.D., Surdeanu, M., Bauer, J., Finkel, J., Bethard, S.J. and McClosky, D. (2014). The Stanford CoreNLP Natural Language Processing Toolkit. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pp. 55-60.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "TextRank: Bringing Order into Text",
"authors": [
{
"first": "R",
"middle": [],
"last": "Mihalcea",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Tarau",
"suffix": ""
}
],
"year": 2004,
"venue": "Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing EMNLP",
"volume": "",
"issue": "",
"pages": "404--411",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mihalcea, R., Tarau, P. (2004). TextRank: Bringing Order into Text. In Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing EMNLP 2004, pp 404-411.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "The Reference Corpus of Contemporary Romanian Language (CoRoLa)",
"authors": [
{
"first": "B",
"middle": [
"V"
],
"last": "Mititelu",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Tufi\u0219",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Irimia",
"suffix": ""
}
],
"year": 2018,
"venue": "Proceedings of the 11th Language Resources and Evaluation Conference -LREC'18",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mititelu, B.V., Tufi\u0219, D. and Irimia, E. (2018). The Reference Corpus of Contemporary Romanian Language (CoRoLa). In Proceedings of the 11th Language Resources and Evaluation Conference - LREC'18, Miyazaki, Japan, European Language Resources Association (ELRA).",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Terminology Extraction: An Analysis of Linguistic and Statistical Approaches",
"authors": [
{
"first": "M",
"middle": [
"T"
],
"last": "Pazienza",
"suffix": ""
},
{
"first": "M",
"middle": [],
"last": "Pennacchiotti",
"suffix": ""
},
{
"first": "F",
"middle": [
"M ;"
],
"last": "Zanzotto",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Tufi\u0219",
"suffix": ""
}
],
"year": 2005,
"venue": "Knowledge Mining. Studies in Fuzziness and Soft Computing",
"volume": "185",
"issue": "",
"pages": "403--409",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Pazienza M.T., Pennacchiotti M. and Zanzotto F.M. (2005). Terminology Extraction: An Analysis of Linguistic and Statistical Approaches. In: Sirmakessis S. (eds) Knowledge Mining. Studies in Fuzziness and Soft Computing, vol 185. Springer, Berlin, Heidelberg P\u0103i\u0219, V., Tufi\u0219, D. (2018). Computing distributed representations of words using the COROLA corpus. In Proceedings of the Romanian Academy, Series A, Volume 19, Number 2/2018, pp. 403-409.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Integration of Romanian NLP tools into the RELATE platform",
"authors": [
{
"first": "V",
"middle": [],
"last": "P\u0103i\u0219",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Tufi\u0219",
"suffix": ""
},
{
"first": "R",
"middle": [],
"last": "Ion",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the International Conference on Linguistic Resources and Tools for Processing Romanian Language -CONSILR 2019",
"volume": "",
"issue": "",
"pages": "181--192",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "P\u0103i\u0219, V., Tufi\u0219, D. and Ion, R. (2019). Integration of Romanian NLP tools into the RELATE platform. In Proceedings of the International Conference on Linguistic Resources and Tools for Processing Romanian Language -CONSILR 2019, pages 181-192.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "TermEval 2020: Shared Task on Automatic Term Extraction Using the Annotated Corpora for Term Extraction Research (ACTER) Dataset",
"authors": [
{
"first": "A",
"middle": [],
"last": "Rigouts Terryn",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Drouin",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Hoste",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Lefever",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of CompuTerm",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rigouts Terryn, A., Drouin, P., Hoste, V., & Lefever, E. (2020). TermEval 2020: Shared Task on Automatic Term Extraction Using the Annotated Corpora for Term Extraction Research (ACTER) Dataset. In Proceedings of CompuTerm 2020.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Analysing the Impact of Supervised Machine Learning on Automatic Term Extraction: HAMLET vs TermoStat",
"authors": [
{
"first": "A",
"middle": [],
"last": "Rigouts Terryn",
"suffix": ""
},
{
"first": "P",
"middle": [],
"last": "Drouin",
"suffix": ""
},
{
"first": "V",
"middle": [],
"last": "Hoste",
"suffix": ""
},
{
"first": "E",
"middle": [],
"last": "Lefever",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of Recent Advances in Natural Language Processing -RANLP 2019",
"volume": "",
"issue": "",
"pages": "1012--1021",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rigouts Terryn, A., Drouin, P., Hoste, V., & Lefever, E. (2019). Analysing the Impact of Supervised Machine Learning on Automatic Term Extraction: HAMLET vs TermoStat. In Proceedings of Recent Advances in Natural Language Processing -RANLP 2019, pages 1012-1021, Varna, Bulgaria, Sep 2-4, 2019.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Automatic keyword extraction from individual documents",
"authors": [
{
"first": "S",
"middle": [],
"last": "Rose",
"suffix": ""
},
{
"first": "D",
"middle": [],
"last": "Engel",
"suffix": ""
},
{
"first": "N",
"middle": [],
"last": "Cramer",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Cowley",
"suffix": ""
}
],
"year": 2010,
"venue": "Text mining: applications and theory",
"volume": "1",
"issue": "",
"pages": "1--20",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Rose, S., Engel, D., Cramer, N., & Cowley, W. (2010). Automatic keyword extraction from individual documents. Text mining: applications and theory, 1, 1- 20.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "A statistical interpretation of term specificity and its application in retrieval",
"authors": [
{
"first": "Sparck",
"middle": [],
"last": "Jones",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "",
"suffix": ""
}
],
"year": 1972,
"venue": "Journal of Documentation",
"volume": "28",
"issue": "",
"pages": "11--21",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sparck Jones, K. (1972). A statistical interpretation of term specificity and its application in retrieval. Journal of Documentation, 28:11-21.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"uris": null,
"num": null,
"text": "RACAI's term extraction system architecture that participated in TermEval 2020",
"type_str": "figure"
},
"TABREF1": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": "Statistics regarding the English sections of the corpus"
},
"TABREF3": {
"type_str": "table",
"html": null,
"content": "<table><tr><td colspan=\"4\">: Precision, Recall, F1 measures for tested</td></tr><tr><td colspan=\"2\">algorithms on the \"equi\" section</td><td/><td/></tr><tr><td/><td>P%</td><td>R%</td><td>F1%</td></tr><tr><td>TFIDF 1W</td><td>16.02</td><td>27.29</td><td>20.19</td></tr><tr><td>TFIDF</td><td>7.81</td><td>18.65</td><td>11.01</td></tr><tr><td>RAKE 1W</td><td>16.80</td><td>75.30</td><td>27.47</td></tr><tr><td>RAKE</td><td>12.95</td><td>65.08</td><td>21.60</td></tr><tr><td>YAKE 1W</td><td>30.94</td><td>8.57</td><td>13.42</td></tr><tr><td>YAKE</td><td>11.81</td><td>9.88</td><td>10.76</td></tr><tr><td>TRANK 1W</td><td>17.67</td><td>39.24</td><td>24.37</td></tr><tr><td>TRANK</td><td>17.05</td><td>18.40</td><td>17.70</td></tr></table>",
"num": null,
"text": ""
},
"TABREF4": {
"type_str": "table",
"html": null,
"content": "<table><tr><td colspan=\"4\">: Precision, Recall, F1 measures for tested</td></tr><tr><td colspan=\"2\">algorithms on the \"corp\" section</td><td/><td/></tr><tr><td/><td>P%</td><td>R%</td><td>F1%</td></tr><tr><td>TFIDF 1W</td><td>17.30</td><td>19.96</td><td>18.54</td></tr><tr><td>TFIDF</td><td>13.18</td><td>11.60</td><td>12.34</td></tr><tr><td>RAKE 1W</td><td>13.62</td><td>58.13</td><td>22.07</td></tr><tr><td>RAKE</td><td>13.90</td><td>63.17</td><td>22.79</td></tr><tr><td>YAKE 1W</td><td>64.29</td><td>3.18</td><td>6.06</td></tr><tr><td>YAKE</td><td>12.37</td><td>3.91</td><td>5.94</td></tr><tr><td>TRANK 1W</td><td>14.57</td><td>34.81</td><td>20.54</td></tr><tr><td>TRANK</td><td>14.11</td><td>13.62</td><td>13.86</td></tr></table>",
"num": null,
"text": ""
},
"TABREF5": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": ""
},
"TABREF6": {
"type_str": "table",
"html": null,
"content": "<table><tr><td/><td/><td>.</td><td/></tr><tr><td/><td>P%</td><td>R%</td><td>F1%</td></tr><tr><td>Equi 1W</td><td>21.28</td><td>87.56</td><td>34.24</td></tr><tr><td>Equi</td><td>7.96</td><td>86.22</td><td>14.57</td></tr><tr><td>Corp 1W</td><td>15.61</td><td>91.43</td><td>26.66</td></tr><tr><td>Corp</td><td>4.85</td><td>89.86</td><td>9.19</td></tr><tr><td>Wind 1W</td><td>13.37</td><td>89.93</td><td>23.28</td></tr><tr><td>Wind</td><td>5.53</td><td>88.33</td><td>10.41</td></tr></table>",
"num": null,
"text": ""
},
"TABREF7": {
"type_str": "table",
"html": null,
"content": "<table/>",
"num": null,
"text": "Precision, Recall, F1 measures for the PLEARN algorithm on the training parts of the corpus"
}
}
}
} |