File size: 104,431 Bytes
6fa4bc9 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T02:10:10.121866Z"
},
"title": "Semi-Supervised Cleansing of Web Argument Corpora",
"authors": [
{
"first": "Jonas",
"middle": [],
"last": "Dorsch",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Webis Group Bauhaus-Universit\u00e4t Weimar Weimar",
"location": {
"country": "Germany"
}
},
"email": "jonas.dorsch@uni-weimar.de"
},
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Paderborn University",
"location": {
"settlement": "Paderborn",
"country": "Germany"
}
},
"email": "henningw@upb.de"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Debate portals and similar web platforms constitute one of the main text sources in computational argumentation research and its applications. While the corpora built upon these sources are rich of argumentatively relevant content and structure, they also include text that is irrelevant, or even detrimental, to their purpose. In this paper, we present a precision-oriented approach to detecting such irrelevant text in a semi-supervised way. Given a few seed examples, the approach automatically learns basic lexical patterns of relevance and irrelevance and then incrementally bootstraps new patterns from sentences matching the patterns. In the existing args.me corpus with 400k argumentative texts, our approach detects almost 87k irrelevant sentences, at a precision of 0.97 according to manual evaluation. With low effort, the approach can be adapted to other web argument corpora, providing a generic way to improve corpus quality.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Debate portals and similar web platforms constitute one of the main text sources in computational argumentation research and its applications. While the corpora built upon these sources are rich of argumentatively relevant content and structure, they also include text that is irrelevant, or even detrimental, to their purpose. In this paper, we present a precision-oriented approach to detecting such irrelevant text in a semi-supervised way. Given a few seed examples, the approach automatically learns basic lexical patterns of relevance and irrelevance and then incrementally bootstraps new patterns from sentences matching the patterns. In the existing args.me corpus with 400k argumentative texts, our approach detects almost 87k irrelevant sentences, at a precision of 0.97 according to manual evaluation. With low effort, the approach can be adapted to other web argument corpora, providing a generic way to improve corpus quality.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Computational argumentation research lays the ground for applications that support opinion formation, including argument search engines (Wachsmuth et al., 2017b) , collective deliberation (Uszkoreit et al., 2017) , and debating technologies (Toledo et al., 2019) . Such applications rely on large pools of up-to-date arguments, which can hardly be found anywere but on the web. One of the most used web argument sources are debate portals where people jointly collect arguments or debate each other on defined issues. Debate portals, and similar web platforms, are rich of argumentatively relevant content and structure, including arguments as well as facts, background information, and similar. This enables researchers to crawl large-scale argument corpora in a distantly-supervised manner (Al-Khatib et al., 2016) .",
"cite_spans": [
{
"start": 136,
"end": 161,
"text": "(Wachsmuth et al., 2017b)",
"ref_id": "BIBREF21"
},
{
"start": 188,
"end": 212,
"text": "(Uszkoreit et al., 2017)",
"ref_id": "BIBREF19"
},
{
"start": 241,
"end": 262,
"text": "(Toledo et al., 2019)",
"ref_id": "BIBREF18"
},
{
"start": 792,
"end": 816,
"text": "(Al-Khatib et al., 2016)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "However, the texts found on debate portals also comprise debate-specific language and boilerplate text that is likely to be irrelevant, if not even detrimental, to the mentioned applications. In the text in Figure 1 , for instance, the author defines the debated issue (sentence #2), states a thesis (#3-5), and presents two arguments (#6-8, #9-13) -all of which can be considered argumentatively relevant. In contrast, sentences #1, #14, and #15 add nothing of importance, merely making meta-comments and expressing gratitude. In other cases, irrelevant text includes salutations, insults, purely rhetorical moves, and spam. As detailed in Section 2, finding such text differs from finding non-argumentative text segments, since the latter may still be relevant as context for the argumentative segments, as in the case of sentence #2 in Figure 1 . Many existing approaches relying on debate portals do not clean the crawled arguments from irrelevant text. Until now, for example, the argument search engine args.me (Wachsmuth et al., 2017b) has just returned the full shown text as one pro argument for the query \"gay marriage\". This at least harms user experience, and it might even corrupt the support of opinion formation in some cases.",
"cite_spans": [
{
"start": 1017,
"end": 1042,
"text": "(Wachsmuth et al., 2017b)",
"ref_id": "BIBREF21"
}
],
"ref_spans": [
{
"start": 207,
"end": 215,
"text": "Figure 1",
"ref_id": null
},
{
"start": 839,
"end": 847,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper, we study how to find irrelevant text in web arguments such as those from debate portals automatically, in order to clean respective corpora on this basis. In particular, we develop a semi-supervised learning approach that aims to detect as many irrelevant sentences as possible with very high precision, i.e., hardly any relevant sentence should be classified as irrelevant (Section 3). Given a seed set of sentences, the approach learns basic lexical n-gram patterns that frequently match text in either relevant or irrelevant I would like to thank Brainmaster for accepting this debate.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Gay marriage is basically the marriage between two individuals of the same gender, I trust my opponent will have no problem with this definition. I will be arguing for gay marriage, and that it should be legal. I will be arguing that everything that does not physically harm other individuals should be legalized, gay marriage is one of these things. I will also be arguing that by banning the gay marriage we have gone against human rights.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "C1: Gay marriage does not physically harm other individuals in any way shape or form therefore, it should be legal. A marriage is a union between two individuals that love eachother, and it basically only effects these two individuals. If it is banned then it is hurting people, and if it is legalized then it isn't hurting anyone.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "C2: Banning gay marriage is against human rights. Every person is born with the equal human rights which are life, liberty, and the pursuit of happiness. Yet, banning gay marriage goes against to of these fundamental rights. How can someone pursue happiness when they can't marry the one they love? How can someone have liberty when they are not allowed to marry the one they love.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "I await my opponent's response. Figure 1 : Example text taken from a debate portal. Sentences #1, #14, and #15 can be considered irrelevant to the arguments made by the author. Our approach learns basic lexical patterns to detect such sentences, here shown bold and underlined. Italicized phrases indicate patterns in sentences learned to be relevant. sentences, and it keeps all patterns with some minimum precision (estimated on all matching sentences). Based on all matching sentences in a given corpus, it then bootstraps new patterns, revises previous ones, and incrementally repeats the process. The final set of irrelevance patterns is used to cleanse the corpus.",
"cite_spans": [],
"ref_spans": [
{
"start": 32,
"end": 40,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We analyze our approach on the args.me corpus (Ajjour et al., 2019) , consisting of 387,606 arguments from four debate portals, more than any other available corpus to our knowledge (Section 4). Exploring different types of lexical patterns, we find that word n-grams ignoring stopwords serve best to distinguish relevant from irrelevant sentences. From the most frequent such n-grams, we manually select a set of seed sentences. Then, we run the bootstrapping process, analyze the patterns found by the approach over its different iterations, and evaluate its precision both in an automatic way and in a manual annotation study with three human annotators on 600 sentences (Section 5). At a Fleiss' \u03ba agreement of 0.50, our approach detects irrelevant sentences with a precision of 0.97, in total 86,916 of them in 68,814 arguments from the args.me corpus. We provide a cleaned version of the corpus to the community. 1 Finally, we discuss how to adopt our approach to improve the quality of web argument corpora, beyond the one studied (Section 6). Altogether, the contribution of this paper is three-fold:",
"cite_spans": [
{
"start": 46,
"end": 67,
"text": "(Ajjour et al., 2019)",
"ref_id": "BIBREF3"
},
{
"start": 919,
"end": 920,
"text": "1",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Vote pro!",
"sec_num": null
},
{
"text": "\u2022 A semi-supervised approach to detect argumentatively irrelevant sentences in web arguments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Vote pro!",
"sec_num": null
},
{
"text": "\u2022 Several common lexical patterns of relevance and irrelevance in web arguments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Vote pro!",
"sec_num": null
},
{
"text": "\u2022 A cleaned version of the largest available argument corpus, with notably less irrelevant text.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Vote pro!",
"sec_num": null
},
{
"text": "Initially, research on tasks such as argument mining has largely been carried out on small, well-curated collections of texts, including Wikipedia articles (Aharoni et al., 2014) , student essays (Stab and Gurevych, 2014) , pure arguments (Peldszus and Stede, 2015) , and presidential debates (Lawrence and Reed, 2017) . Major real-world applications of computational argumentation, however, need to scale up to web contexts to fulfill their purpose. This includes search engines that oppose pro and con arguments on controversial issues (Wachsmuth et al., 2017b) , technologies that debate humans (Toledo et al., 2019) , and more.",
"cite_spans": [
{
"start": 156,
"end": 178,
"text": "(Aharoni et al., 2014)",
"ref_id": "BIBREF1"
},
{
"start": 196,
"end": 221,
"text": "(Stab and Gurevych, 2014)",
"ref_id": "BIBREF17"
},
{
"start": 239,
"end": 265,
"text": "(Peldszus and Stede, 2015)",
"ref_id": "BIBREF16"
},
{
"start": 293,
"end": 318,
"text": "(Lawrence and Reed, 2017)",
"ref_id": "BIBREF14"
},
{
"start": 538,
"end": 563,
"text": "(Wachsmuth et al., 2017b)",
"ref_id": "BIBREF21"
},
{
"start": 598,
"end": 619,
"text": "(Toledo et al., 2019)",
"ref_id": "BIBREF18"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "To obtain web arguments, many works have relied on crawled debate portals and similar web platforms, often in a distant-supervision manner where argumentative structure and similar annotations are directly derived from available meta-information (Al-Khatib et al., 2016) . Corpora have been built in such a way based on several debate portals, including 4forums.com (Walker et al., 2012) , idebate.org (Cabrio and Villata, 2012) , createdebate.com (Habernal and Gurevych, 2016) , debate.org (Durmus and Cardie, 2019) , and reddit.com/r/changemyview (Egawa et al., 2020) . Naturally, less curation of the acquired web texts comes at the cost of more noise, which in turn calls for a cleansing of the resulting corpus.",
"cite_spans": [
{
"start": 246,
"end": 270,
"text": "(Al-Khatib et al., 2016)",
"ref_id": "BIBREF4"
},
{
"start": 366,
"end": 387,
"text": "(Walker et al., 2012)",
"ref_id": "BIBREF22"
},
{
"start": 402,
"end": 428,
"text": "(Cabrio and Villata, 2012)",
"ref_id": "BIBREF7"
},
{
"start": 448,
"end": 477,
"text": "(Habernal and Gurevych, 2016)",
"ref_id": "BIBREF12"
},
{
"start": 491,
"end": 516,
"text": "(Durmus and Cardie, 2019)",
"ref_id": "BIBREF9"
},
{
"start": 549,
"end": 569,
"text": "(Egawa et al., 2020)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "Cleansing processes are described in several publications on argument corpora, mostly only referring to the acquired annotations though (Habernal and Gurevych, 2016; Toledo et al., 2019; Gretz et al., 2020) . In contrast, the paper at hand targets the cleansing of the corpus texts themselves. Only few works describe respective cleansing steps in detail. Among these, Al-Khatib et al. (2016) deleted special symbols and debate-specific phrases such as \"this house\" from crawled arguments, and Habernal and Gurevych (2017) removed quotations of previous posts in debate posts. Wachsmuth et al. (2017b) discarded certain types of noisy instances completely for the argument search engine args.me, but the texts in the original associated corpus (Ajjour et al., 2019) still contain much irrelevant text, as our experiments will reveal. Applying our approach has led to an improved version of that corpus.",
"cite_spans": [
{
"start": 136,
"end": 165,
"text": "(Habernal and Gurevych, 2016;",
"ref_id": "BIBREF12"
},
{
"start": 166,
"end": 186,
"text": "Toledo et al., 2019;",
"ref_id": "BIBREF18"
},
{
"start": 187,
"end": 206,
"text": "Gretz et al., 2020)",
"ref_id": null
},
{
"start": 494,
"end": 522,
"text": "Habernal and Gurevych (2017)",
"ref_id": "BIBREF13"
},
{
"start": 577,
"end": 601,
"text": "Wachsmuth et al. (2017b)",
"ref_id": "BIBREF21"
},
{
"start": 744,
"end": 765,
"text": "(Ajjour et al., 2019)",
"ref_id": "BIBREF3"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "In this paper, we introduce a semi-supervised learning approach for corpus cleansing. In general, we follow the bootstrapping idea of successful pattern mining methods, such as DIPRE (Brin, 1998), Snowball (Agichtein and Gravano, 2000) , and Espresso (Pantel and Pennacchiotti, 2006) . While these methods aim at semantically relevant information, we distinguish pragmatically relevant from irrelevant text within an author's argumentative discourse. We are not aware of any other approach in this direction.",
"cite_spans": [
{
"start": 206,
"end": 235,
"text": "(Agichtein and Gravano, 2000)",
"ref_id": "BIBREF0"
},
{
"start": 251,
"end": 283,
"text": "(Pantel and Pennacchiotti, 2006)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "It is noteworthy in this regard that the cleansing task at hand differs notably from the unit segmentation of argumentative texts (Ajjour et al., 2017) . While all argumentative units match the notion of relevance considered here (defined in Section 3), also non-argumentative units may be seen as relevant, if they give facts, definitions, or other background information serving as context for the argumentative units. As such, our notion of relevance relates to the local relevance with respect to some conclusion rather than the global relevance of an argumentative statement in the discussion of an issue (Wachsmuth et al., 2017a) .",
"cite_spans": [
{
"start": 130,
"end": 151,
"text": "(Ajjour et al., 2017)",
"ref_id": "BIBREF2"
},
{
"start": 610,
"end": 635,
"text": "(Wachsmuth et al., 2017a)",
"ref_id": "BIBREF20"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Related Work",
"sec_num": "2"
},
{
"text": "This section presents our semi-supervised learning approach to detecting irrelevant text in web arguments as well as to clean a respective corpus on this basis. The approach aims to find as many irrelevant text units as possible at an estimated precision beyond a threshold \u03c4 (in Section 5, we use \u03c4 = 0.95). To this end, it learns linguistic patterns that occur often in irrelevant units and rarely in relevant units (or vice versa). Later, we consider each sentence as one unit, but other granularities would work in principle, too. Figure 2 gives an overview of the three main stages of the approach, each of which will be detailed below:",
"cite_spans": [],
"ref_spans": [
{
"start": 535,
"end": 543,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Approach",
"sec_num": "3"
},
{
"text": "(a) Seed Pattern Selection. Given a corpus as input, a pool of common linguistic patterns is mined from its units, from which seed patterns indicating irrelevance and relevance are selected manually.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "3"
},
{
"text": "(b) Pattern Bootstrapping. All units matching any seed irrelevance (relevance) pattern are retrieved, new candidate patterns are mined from the units and added to the pool. Then, only high-precision irrelevance (relevance) patterns are kept in the pool, i.e., those found nearly only in irrelevant (relevant) units. This process is repeated until no new patterns are found or k iterations have passed.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "3"
},
{
"text": "(c) Corpus Cleansing. The final pool of irrelevance patterns is used to automatically remove irrelevant units from the corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "3"
},
{
"text": "It is important to see that the relevance patterns are eventually not used for the actual cleansing. They serve to distinguish relevant from irrelevant units only, thereby aiding the identification high-precision irrelevance patterns.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "3"
},
{
"text": "While we have designed our approach for web arguments in particular, notice that the outlined processed is largely generic and could easily be transferred to other cleansing tasks where relevant and irrelevant units can be distinguished. What makes our approach specific to web arguments is what we mean by argumentative relevance and irrelevance.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Approach",
"sec_num": "3"
},
{
"text": "We consider relevance here from the perspective of using the individual arguments in a corpus for empirical analysis of how people argue or for applications such as argument search and debating technologies. For such use cases, portal-specific debate structure emerging from sequences of arguments as well as purely rhetorical moves related to the underlying debates are not of interest. We thus define irrelevance as follows: Argumentative Irrelevance. A unit of a web argument is said to be irrelevant, if and only if it does not represent any claim, evidence, fact, background information, or similar statement related to the issue discussed by the author of the text. Examples of irrelevant units include meta-comments on a debate, salutations, expressions of gratitude, personal insults, purely rhetorical moves, and spam.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Argumentative Relevance and Irrelevance",
"sec_num": "3.1"
},
{
"text": "Any unit not matching the definition is considered to be relevant. While we could have also defined argumentative relevance instead, we decided to focus on irrelevant units, since they constitute the target concept to be detected. In other words, given that we target argument corpora, we expect irrelevant units to be the exception rather than the default. An estimation of the proportion of irrelevant units for the data processed in our experiments follows in Section 4.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Argumentative Relevance and Irrelevance",
"sec_num": "3.1"
},
{
"text": "The goal of stage (a) is to acquire a pool of linguistic patterns matching text units that can be considered either irrelevant or relevant. The set of all units matching any of these seed patterns then represents the ground-truth data that the pattern bootstrapping starts from. The selection of seed patterns is the only step that requires some level of supervision within our approach. To minimize manual effort, we propose to tackle the selection semi-automatically, i.e., we first mine the most promising candidate patterns automatically from sample data (we use a random 10% sample of the given corpus in Section 5). Then, we manually classify a subset of them to be seed patterns either of irrelevance or of relevance. To do so, however, we need to first define what is considered to be a candidate pattern.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Seed Pattern Selection",
"sec_num": "3.2"
},
{
"text": "Candidate Patterns. In general, any type of linguistic pattern may be mined from corpus texts, for which respective mining methods are available. Since we expect the given notion of relevance to be largely assessable based on a unit's words only, we restrict our view to basic lexical patterns here. For simplicity, we just look for n-grams, but we explore four types of patterns that emerge from making two choices:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Seed Pattern Selection",
"sec_num": "3.2"
},
{
"text": "\u2022 Counts vs. TF-IDF. In case of counts, we simply see the m most frequent n-grams as candidates for each n. In case of TF-IDF, we take those n-grams with the highest TF-IDF score in the sample data (each unit being one document). In our experiments, we use m = 100 and n \u2208 {1, . . . , 5}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Seed Pattern Selection",
"sec_num": "3.2"
},
{
"text": "\u2022 W/ stopwords vs. w/o stopwords. We determine either n-grams based on the full unit texts (w/ stopwords) or we apply stopword removal before (w/o stopwords).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Seed Pattern Selection",
"sec_num": "3.2"
},
{
"text": "Since high TF-IDF scores usually indicate content, respective patterns are likely to be more useful for relevant than irrelevant sentences. Whether they outperform count-based patterns there is hard to predict, though. In Section 5, we compare the four pattern types against each other. Given all m candidates of the preferred pattern type (say, Counts w/o stopwords) for each n, the authors of this paper then manually agree for each candidate on whether to select it as an irrelevance pattern, a relevance pattern, or neither.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Seed Pattern Selection",
"sec_num": "3.2"
},
{
"text": "The goal of stage (b) is to incrementally extend the pool of irrelevance and relevance patterns using bootstrapping, i.e., by deriving new patterns from units matching the current patterns in the pool. This fully automatic process continues until no new patterns are found anymore or until a maximum number k of iterations has passed, e.g., if running time is a factor (in Section 5, we continue until the end).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pattern Bootstrapping",
"sec_num": "3.3"
},
{
"text": "In particular, the first step is to retrieve the sets of all units matching any irrelevance patterns and of all units matching any relevance pattern from the corpus. 2 As sketched in Figure 2 , these unit sets are used for two purposes: First, new candidate irrelevance (relevance) patterns are mined from the set of irrelevant (relevant) units and added to the pattern pool. Second, only those patterns are filtered and kept in the pool that indicate an irrelevant (relevant) unit with an estimated precision p \u2265 \u03c4 . We estimate p as follows:",
"cite_spans": [],
"ref_spans": [
{
"start": 183,
"end": 191,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Pattern Bootstrapping",
"sec_num": "3.3"
},
{
"text": "Estimated Precision. Let tp be the number of all retrieved irrelevant (relevant) units that matches a specific irrelevance (relevance) pattern, and let f p be the number of all relevant (irrelevant) units matching this pattern. Then the precision of the pattern is estimated as p = tp / (tp + f p).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pattern Bootstrapping",
"sec_num": "3.3"
},
{
"text": "For the mining step, one parameter to decide upon is the minimum frequency of a pattern to consider it a candidate. We suggest to derive this parameter's value from the seed pattern frequencies. For example, if all seed patterns have at least 20 matches in the sample, and the full corpus has 10 times the sample size, then a reasonable value may be 20 \u2022 10 = 200. For the filtering step, it is favorable that the sizes of the two unit sets remain balanced, because imbalanced sizes decrease the comparability of the values tp and f p. We therefore suggest to adjust the minimum numbers based on the estimated proportion of irrelevant units. For example, if there are about 10 times as many relevant as irrelevant units, reasonable values may be 200 for irrelevance and 200 \u2022 10 = 2000 for relevance (the numbers given here exemplarily are those we use in Sections 4 and 5). An alternative is to test and adjust these parameters empirically.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pattern Bootstrapping",
"sec_num": "3.3"
},
{
"text": "An important characteristic of the outlined bootstrapping process is that patterns added to the pool in previous iterations may be removed later from the pool again. This is because the sets of retrieved relevant and irrelevant units change during the process, which in turn may change the precision estimations of the patterns. This can be understood as an internal revision mechanism of our approach that optimizes the precision of the final pool. We see the effect of this mechanism in our experiments in Section 5. 3",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pattern Bootstrapping",
"sec_num": "3.3"
},
{
"text": "The goal of stage (c) is to actually clean the given corpus, based on the final pool of irrelevance patterns. Relevance patterns play no role anymore in this stage; they are used only before, to be able to help identify irrelevance patterns with high precision, as described.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Corpus Cleansing",
"sec_num": "3.4"
},
{
"text": "A simple cleansing way would be to just remove all units from the corpus that match any irrelevance patterns. Instead, however, we suggest to restrict the removal to only those irrelevant units before the first and after the last relevant unit. As long as only units are removed that are actually irrelevant, we thereby avoid to negatively affect the coherence of arguments. Moreover, as for the example of Figure 1 , we will see below that most irrelevant units are indeed found in the beginning and ending of texts, i.e., the suggested restriction reduces recall to some extent only. Notice that this does not mean that most units in the beginning and ending are irrelevant; in line with our discussions above, we expect the majority of texts to contain no irrelevant unit at all. The following section supports that this is true for the corpus at hand.",
"cite_spans": [],
"ref_spans": [
{
"start": 407,
"end": 415,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Corpus Cleansing",
"sec_num": "3.4"
},
{
"text": "The presented approach targets argumentative language of varying quality, as often observed in web-based corpora. Below, we assess its impact on the args.me corpus (Ajjour et al., 2019) , which is to our knowledge the largest available argument corpus to this date, about 7.3 GB in file size. The corpus represents the database underlying the argument search engine args.me (Wachsmuth et al., 2017b Table 1 : The top n-gram patterns agreed upon to indicate relevant and irrelevant sentences respectively, for each evaluated pattern type, along with their score (count or TF-IDF) in the 10% sample of the args.me corpus. We left out spam patterns, such as \"kfc ... kfc\", as they would have shadowed most other patterns. Based on the full lists (see supplementary material), we decided to use the type Counts w/o Stopwords.",
"cite_spans": [
{
"start": 164,
"end": 185,
"text": "(Ajjour et al., 2019)",
"ref_id": "BIBREF3"
},
{
"start": 374,
"end": 398,
"text": "(Wachsmuth et al., 2017b",
"ref_id": "BIBREF21"
}
],
"ref_spans": [
{
"start": 399,
"end": 406,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Data",
"sec_num": "4"
},
{
"text": "arguments that were mined from four debate portals using distant supervision: debate.org, debatewise.org, idebate.org, and debatepedia.org. Each argument consists of a mostly very short conclusion as well as a mostly notably longer premise, the latter containing the actual argumentative text. In total, the corpus spans around seven million sentences. We see each sentence as one unit in our approach. Many texts in the args.me corpus include sentences that are irrelevant to the actual argument, such as the example in Figure 1 . Needless to say, no ground-truth information on irrelevance is given, though. For a rough estimation of the proportion of irrelevant sentences, we conducted a pilot study where the two authors of this paper independently decided about the relevance of a set of sentences, following the definition in Section 3. In particular, we considered a corpus sample used previously by Alshomary et al. (2020) , which contains the top five pro and the top five con arguments each for the top 10 queries.",
"cite_spans": [
{
"start": 907,
"end": 930,
"text": "Alshomary et al. (2020)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 521,
"end": 529,
"text": "Figure 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Data",
"sec_num": "4"
},
{
"text": "From the 1294 sentences in the 100 sample arguments, one of us classified 147 (11.3%) to be irrelevant, the other one 139 (10.7%). In terms of Cohen's \u03ba, we had a substantial inter-annotator agreement of 0.75. In total, 175 sentences (13.5%) were seen as irrelevant by either of us, 111 (8.5%) by both. Since we believe that, in doubt, a sentence should be deemed relevant, we take 8.5% as our estimation. In the whole corpus, we thus expect around 600,000 sentences to be irrelevant. The 111 sentences come from only 39 of the 100 arguments. Assuming this number is representative, about 150k arguments in the corpus should contain irrelevant sentences. In the following experiments, these numbers will give us a rough idea of the recall of our approach. There, we use a random 10% sample of all corpus arguments for the seed pattern selection, and the whole corpus for all subsequent steps.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Data",
"sec_num": "4"
},
{
"text": "We now report on the step-by-step application of our approach from Section 3 to the corpus from Section 4 and on the manual evaluation of the obtained results. The goal was to assess the impact of the approach on the quality of web-based argument corpora. We hypothesized that the approach is able to detect a large number of irrelevant sentences with a precision as high as its internal precision threshold \u03c4 . 4 The full lists of positive and negative and seed patterns used for each n-gram type, along with the number of different sentences they match in the corpus (in parentheses), ordered by number of matches.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation",
"sec_num": "5"
},
{
"text": "To learn what pattern type is best to detect irrelevant sentences, we compared all four candidates emerging from the two choices discussed in Section 3 (Counts vs. TF-IDF, w/ or w/o stopwords). For each type, we retrieved the top 100 n-grams, n \u2208 {1, . . . , 5}, covering a large variety of issues debated in the underlying arguments. Then, the two authors of this paper both judged all 2000 resulting patterns as to whether they likely indicate always irrelevant sentences or always relevant ones. Based on the patterns that we both agreed upon, the most promising type was chosen for the seed patterns. Exemplarily, Table 1 lists the top 1-to 5-gram of each pattern type that indicate relevance or irrelevance respectively. We left out spam patterns such as \"wonyou wonyou wonyou\" and \"kfc kfc\", though, as they would limit insights, dominating the top positions; the full lists for each pattern type are given in the supplementary material. For both TF-IDF pattern types, we find the relevance patterns to clearly serve their purpose, relating to the content of arguments. Many such patterns are found in the full lists. However, rarely any TF-IDF pattern seemed to reliably indicate irrelevance. This matches the intuition that phrases with high TF-IDF scores are specific to a document's content rather than reflecting general language. In contrast, the two Counts pattern types yielded several irrelevance patterns, as the table demonstrates. We decided for Counts w/o Stopwords, since it produced patterns that clarified many cases which Counts w/ Stopwords left ambiguous. For example, \"would like thank opponent\" reveals irrelevance knowing the source debate portals (here, debate.org), whereas respective patterns with stopwords (\"would like to thank\", \"like to thank my\") leaves more doubts regarding the irrelevance of respective sentences. Table 2 presents the full set of 38 relevance and 17 irrelevance seed patterns for the type Counts w/o Stopwords. A pattern was not included if being redundant, i.e., if it was already covered by a shorter one, e.g., \"first round acceptance\" was covered by \"first round\". We observe that no 1-gram made it into the pool of irrelevance patterns; a single word seems not enough to be sure about irrelevance. As of length 2, however, we judged several patterns to be sufficiently reliable indicators of irrelevance, the most frequent ones occurring over 10,000 times in the corpus, namely, \"first round\" and \"thank opponent\".",
"cite_spans": [],
"ref_spans": [
{
"start": 618,
"end": 625,
"text": "Table 1",
"ref_id": null
},
{
"start": 1853,
"end": 1860,
"text": "Table 2",
"ref_id": "TABREF3"
}
],
"eq_spans": [],
"section": "Insights into Seed Pattern Selection",
"sec_num": "5.1"
},
{
"text": "As indicated in Section 3, we set \u03c4 to 0.95, kept all mined relevance patterns with at least 2000 matches as candidates and all mined irrelevance patterns with at least 200 matches. Given the seed patterns, we then ran the bootstrapping process until no new pattern was found anymore, which happened in iteration 6. On a standard computer (Intel Core i7, 2.7 GHz, 16 GB RAM), the whole process took about two hours. Table 3 shows key statistics for each iteration (and the seed pattern selection). In case of the relevance Table 3 : Counts of relevance and irrelevance patterns, counts of different sentences they match, their automatically estimated mean precision, and their manually evaluated mean precision (majority agreement, full agreement in parentheses) in each iteration of our approach. The last row shows the results at the end.",
"cite_spans": [],
"ref_spans": [
{
"start": 416,
"end": 423,
"text": "Table 3",
"ref_id": null
},
{
"start": 523,
"end": 530,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Insights into Pattern Bootstrapping",
"sec_num": "5.2"
},
{
"text": "patterns, the 38 seed patterns already match more than 600k different sentences, with a mean estimated precision of 1.00, i.e., they virtually never matched any sentence retrieved for the seed irrelevance patterns. Already in iteration 2, the revision effect discussed in Section 3 starts: 57 relevant sentences were removed there, because they also matched newly mined irrelevance patterns. Still, the set of relevance patterns remained stable, and this behavior continued in subsequent iterations. For the irrelevance patterns, we observe a monotonous growth of the pattern pool in the first five iterations, with more than 10k different sentences being detected as irrelevant in iterations 1-5 in addition to the seed sentences. In total, 122 patterns were found; their mean estimated precision remained at least 0.97 in all iterations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Insights into Pattern Bootstrapping",
"sec_num": "5.2"
},
{
"text": "To analyze the behavior of our approach during the bootstrapping process, we chose a random sample of 600 irrelevant sentences for manual evaluation (found in the supplementary material): 100 matching the seed irrelevance patterns, and 100 each for the irrelevance patterns from the five iterations. Relevant patterns were disregarded, as they are not needed for corpus cleansing. We randomized the ordering of all sentences and gave them independently to three annotators with background on computational argumentation, none being an author of this paper (one master and two PhD students; two male, one female; one each from Europe, the Middle East, and East Asia). We asked the annotators to classify each sentence as relevant or irrelevant, based on the definition from Section 3. The annotators got some intuitive guidelines (see supplementary material) and could ask questions beforehand.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Insights into Pattern Bootstrapping",
"sec_num": "5.2"
},
{
"text": "We observed an inter-annotator agreement of 0.50 in terms of Fleiss' \u03ba, which seems reasonable given that relevance assessment is inherently subjective (Croft et al., 2009) . Given the annotations, we computed the mean precision of our approach in detecting irrelevant sentences for each iteration, once in terms of majority agreement (irrelevance correct if two annotators say so) and once for full agreement (all three say so). The right-most column in Table 3 shows the results, revealing that the majority-agreement precision is perfect until the end of iteration 2. While the next two iterations remain promising, the precision decreases to 0.88 in the final iteration (0.79 under full agreement), suggesting that patterns get worse over time. An early termination may thus be favorable, but the best moment is naturally unknown in practice.",
"cite_spans": [
{
"start": 152,
"end": 172,
"text": "(Croft et al., 2009)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 455,
"end": 462,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Insights into Pattern Bootstrapping",
"sec_num": "5.2"
},
{
"text": "52,849 different sentences are matched by the detected irrelevance patterns eventually, at an overall precision of 0.97. Some of them occur multiple times, resulting in 86,916 irrelevant sentences in total that come from 68,814 arguments. Under the roughly estimated irrelevance proportion from Section 4, the recall would hence be around 0.15 for irrelevant sentences and around 0.46 for arguments with irrelevance sentences. The seed step alone found 71,926 irrelevant sentences in total, i.e., a recall of roughly 0.12. If we consider the seed step as a baseline for the full approach, we see that precision decreases by 3% (1.00 to 0.97), but recall increases by about 20% (0.12 to 0.15). While there is arguably room for optimization, we still conclude that the results support the impact of our approach and, by that, our hypothesis.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Insights into Pattern Bootstrapping",
"sec_num": "5.2"
},
{
"text": "Based on the final pool of 122 irrelevance patterns, we explored the cleansing potential for the given corpus. Figure 3(a) shows a histogram of the corpus texts with a certain number of detected irrelevant sentences. We see that most texts contain one such sentence only, in all but six cases seven or less. These six cases all have more than 30 irrelevant sentences; manual inspection revealed that they all contain spam where the same word sequence repeats itself. In Figure 3(b) , we plot the positions of irrelevant sentences in the corpus texts. As expected, most of them are found in the beginning or the end. Due to our discussed restriction of discarding only these, the final number of sentences removed from the args.me corpus sums up to 53,502 (found in 48,089 arguments). In addition to the original args.me corpus, we now also provide the cleaned corpus version at https://webis.de/data.html#args-me-corpus.",
"cite_spans": [],
"ref_spans": [
{
"start": 111,
"end": 122,
"text": "Figure 3(a)",
"ref_id": "FIGREF2"
},
{
"start": 470,
"end": 481,
"text": "Figure 3(b)",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Insights into Corpus Cleansing",
"sec_num": "5.3"
},
{
"text": "Web-based argument corpora play an important role in computational argumentation research and its applications. Not all text in such corpora is relevant to the arguments, though. In this paper, we have presented an approach that detects irrelevant text units in argumentative texts with low supervision. The approach iteratively bootstraps linguistic patterns of irrelevance and relevance from units matching known patterns. On the 387k arguments in the args.me corpus, the approach detected 87k irrelevant sentences at a precision of 0.97, from which at least 53k can be removed without notably reducing the arguments' coherence. These results demonstrate the potential of our approach to improve corpus quality. Naturally, the approach has limitations. On one hand, the results revealed that, under the employed configuration, a large proportion of detected sentences came from the seed patterns. To obtain good seed patterns, manual effort is needed. On the other hand, the recall of our approach seems not so high, as far as we can estimate from the data inspected. While not all irrelevant units can be captured by the simple patterns we considered, another reason may lie in the restriction that only new candidate patterns are found which occur in sentences matching previous patterns. Particularly patterns that show up only in short units may thus be overlooked, if they are not covered by the seed patterns already. Improvements might, e.g., consider units adjacent to irrelevant units, but this may come at the cost of reduced precision. In this regard, notice that the impact our approach to some extent depends on the availability of a reliable unit boundary detector (say, a sentence splitter), which is not a trivial requirement for noisy web data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "Finally, an arising question may be how complex it is to apply the approach to other than the data processed here. Following our proposed process to obtain frequent candidate seed patterns automatically, the main manual effort boils down to finding reliable seed patterns among these candidates. In our case, this took no more than a few hours, which seems negligible given the potential impact on corpus quality. Besides, only some initial tuning of the approach parameters to the data at hand may be needed. We are thus confident that the approach can be easily adopted to clean other argument corpora (including transcribed corpora with spoken argumentative language) as well as to many other cleansing tasks where the irrelevance of text units can be defined in a measurable way.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "6"
},
{
"text": "Both the original and the cleaned args.me corpus are found at: https://webis.de/data.html#args-me-corpus",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "We include units that match both relevance and irrelevance patterns, since the subsequent filtering step accounts for them. Also, other performance optimizations are useful, such as storing previously found units. We leave them out here for simplicity.3 Depending on what sentences match the patterns, it is theoretically possible that a pattern first belongs to the relevance pool and later to the irrelevance pool (or vice versa). We did not observe notable cases in this regard, though.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Source code and supplementary material can be found here: https://github.com/webis-de/ArgMining-20",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We thank Milad Alshomary, Wei-Fan Chen, and Jana Puschmann for their participation in the manual evaluation, and the anonymous reviewers for their helpful comments. Thank you also to Johannes Kiesel as part of the Webis Group for the technical support and the integration of the results into args.me.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgments",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Snowball: Extracting relations from large plain-text collections",
"authors": [
{
"first": "Eugene",
"middle": [],
"last": "Agichtein",
"suffix": ""
},
{
"first": "Luis",
"middle": [],
"last": "Gravano",
"suffix": ""
}
],
"year": 2000,
"venue": "Proceedings of the Fifth ACM Conference on Digital Libraries, DL '00",
"volume": "",
"issue": "",
"pages": "85--94",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eugene Agichtein and Luis Gravano. 2000. Snowball: Extracting relations from large plain-text collections. In Proceedings of the Fifth ACM Conference on Digital Libraries, DL '00, pages 85-94, New York, NY, USA. Association for Computing Machinery.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "A benchmark dataset for automatic detection of claims and evidence in the context of controversial topics",
"authors": [
{
"first": "Ehud",
"middle": [],
"last": "Aharoni",
"suffix": ""
},
{
"first": "Anatoly",
"middle": [],
"last": "Polnarov",
"suffix": ""
},
{
"first": "Tamar",
"middle": [],
"last": "Lavee",
"suffix": ""
},
{
"first": "Daniel",
"middle": [],
"last": "Hershcovich",
"suffix": ""
},
{
"first": "Ran",
"middle": [],
"last": "Levy",
"suffix": ""
},
{
"first": "Ruty",
"middle": [],
"last": "Rinott",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Gutfreund",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Slonim",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the First Workshop on Argumentation Mining",
"volume": "",
"issue": "",
"pages": "64--68",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ehud Aharoni, Anatoly Polnarov, Tamar Lavee, Daniel Hershcovich, Ran Levy, Ruty Rinott, Dan Gutfreund, and Noam Slonim. 2014. A benchmark dataset for automatic detection of claims and evidence in the context of controversial topics. In Proceedings of the First Workshop on Argumentation Mining, pages 64-68, Baltimore, Maryland, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Unit segmentation of argumentative texts",
"authors": [
{
"first": "Yamen",
"middle": [],
"last": "Ajjour",
"suffix": ""
},
{
"first": "Wei-Fan",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Johannes",
"middle": [],
"last": "Kiesel",
"suffix": ""
},
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 4th Workshop on Argument Mining",
"volume": "",
"issue": "",
"pages": "118--128",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yamen Ajjour, Wei-Fan Chen, Johannes Kiesel, Henning Wachsmuth, and Benno Stein. 2017. Unit segmentation of argumentative texts. In Proceedings of the 4th Workshop on Argument Mining, pages 118-128. Association for Computational Linguistics.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Data acquisition for argument search: The args.me corpus",
"authors": [
{
"first": "Yamen",
"middle": [],
"last": "Ajjour",
"suffix": ""
},
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Johannes",
"middle": [],
"last": "Kiesel",
"suffix": ""
},
{
"first": "Martin",
"middle": [],
"last": "Potthast",
"suffix": ""
},
{
"first": "Matthias",
"middle": [],
"last": "Hagen",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2019,
"venue": "KI 2019: Advances in Artificial Intelligence -42nd German Conference on AI",
"volume": "",
"issue": "",
"pages": "48--59",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yamen Ajjour, Henning Wachsmuth, Johannes Kiesel, Martin Potthast, Matthias Hagen, and Benno Stein. 2019. Data acquisition for argument search: The args.me corpus. In KI 2019: Advances in Artificial Intelligence - 42nd German Conference on AI, Kassel, Germany, September 23-26, 2019, Proceedings, pages 48-59.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Cross-domain mining of argumentative text through distant supervision",
"authors": [
{
"first": "Khalid",
"middle": [],
"last": "Al-Khatib",
"suffix": ""
},
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Matthias",
"middle": [],
"last": "Hagen",
"suffix": ""
},
{
"first": "Jonas",
"middle": [],
"last": "K\u00f6hler",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "",
"issue": "",
"pages": "1395--1404",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Khalid Al-Khatib, Henning Wachsmuth, Matthias Hagen, Jonas K\u00f6hler, and Benno Stein. 2016. Cross-domain mining of argumentative text through distant supervision. In Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1395-1404. Association for Computational Linguistics.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Extractive snippet generation for arguments",
"authors": [
{
"first": "Milad",
"middle": [],
"last": "Alshomary",
"suffix": ""
},
{
"first": "Nick",
"middle": [],
"last": "D\u00fcsterhus",
"suffix": ""
},
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
}
],
"year": 2020,
"venue": "43nd International ACM Conference on Research and Development in Information Retrieval, SIGIR '20",
"volume": "",
"issue": "",
"pages": "1969--1972",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Milad Alshomary, Nick D\u00fcsterhus, and Henning Wachsmuth. 2020. Extractive snippet generation for arguments. In 43nd International ACM Conference on Research and Development in Information Retrieval, SIGIR '20, pages 1969-1972, New York, NY, USA. Association for Computing Machinery.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Extracting patterns and relations from the world wide web",
"authors": [
{
"first": "",
"middle": [],
"last": "Sergey Brin",
"suffix": ""
}
],
"year": 1998,
"venue": "Selected Papers from the International Workshop on The World Wide Web and Databases, WebDB '98",
"volume": "",
"issue": "",
"pages": "172--183",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sergey Brin. 1998. Extracting patterns and relations from the world wide web. In Selected Papers from the In- ternational Workshop on The World Wide Web and Databases, WebDB '98, pages 172-183, Berlin, Heidelberg. Springer-Verlag.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Combining textual entailment and argumentation theory for supporting online debates interactions",
"authors": [
{
"first": "Elena",
"middle": [],
"last": "Cabrio",
"suffix": ""
},
{
"first": "Serena",
"middle": [],
"last": "Villata",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics",
"volume": "2",
"issue": "",
"pages": "208--212",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Elena Cabrio and Serena Villata. 2012. Combining textual entailment and argumentation theory for supporting online debates interactions. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 208-212. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Search Engines: Information Retrieval in Practice",
"authors": [
{
"first": "Bruce",
"middle": [],
"last": "Croft",
"suffix": ""
},
{
"first": "Donald",
"middle": [],
"last": "Metzler",
"suffix": ""
},
{
"first": "Trevor",
"middle": [],
"last": "Strohman",
"suffix": ""
}
],
"year": 2009,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Bruce Croft, Donald Metzler, and Trevor Strohman. 2009. Search Engines: Information Retrieval in Practice. Addison-Wesley, USA, 1st edition.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "A corpus for modeling user and language effects in argumentation on online debating",
"authors": [
{
"first": "Esin",
"middle": [],
"last": "Durmus",
"suffix": ""
},
{
"first": "Claire",
"middle": [],
"last": "Cardie",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "602--607",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Esin Durmus and Claire Cardie. 2019. A corpus for modeling user and language effects in argumentation on online debating. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 602-607, Florence, Italy, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Corpus for modeling user interactions in online persuasive discussions",
"authors": [
{
"first": "Ryo",
"middle": [],
"last": "Egawa",
"suffix": ""
},
{
"first": "Gaku",
"middle": [],
"last": "Morio",
"suffix": ""
},
{
"first": "Katsuhide",
"middle": [],
"last": "Fujita",
"suffix": ""
}
],
"year": 2020,
"venue": "Proceedings of The 12th Language Resources and Evaluation Conference",
"volume": "",
"issue": "",
"pages": "1135--1141",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ryo Egawa, Gaku Morio, and Katsuhide Fujita. 2020. Corpus for modeling user interactions in online persuasive discussions. In Proceedings of The 12th Language Resources and Evaluation Conference, pages 1135-1141, Marseille, France, May. European Language Resources Association.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Ranit Aharonov, and Noam Slonim. 2020. A large-scale dataset for argument quality ranking: Construction and analysis",
"authors": [
{
"first": "Shai",
"middle": [],
"last": "Gretz",
"suffix": ""
},
{
"first": "Roni",
"middle": [],
"last": "Friedman",
"suffix": ""
},
{
"first": "Edo",
"middle": [],
"last": "Cohen-Karlik",
"suffix": ""
},
{
"first": "Assaf",
"middle": [],
"last": "Toledo",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Lahav",
"suffix": ""
}
],
"year": null,
"venue": "Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "7805--7813",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shai Gretz, Roni Friedman, Edo Cohen-Karlik, Assaf Toledo, Dan Lahav, Ranit Aharonov, and Noam Slonim. 2020. A large-scale dataset for argument quality ranking: Construction and analysis. In Proceedings of the Thirty-Fourth AAAI Conference on Artificial Intelligence, pages 7805-7813. AAAI.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Which argument is more convincing? Analyzing and predicting convincingness of web arguments using bidirectional lstm",
"authors": [
{
"first": "Ivan",
"middle": [],
"last": "Habernal",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "1589--1599",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ivan Habernal and Iryna Gurevych. 2016. Which argument is more convincing? Analyzing and predicting con- vincingness of web arguments using bidirectional lstm. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1589-1599. Association for Com- putational Linguistics.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Argumentation mining in user-generated web discourse",
"authors": [
{
"first": "Ivan",
"middle": [],
"last": "Habernal",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2017,
"venue": "Computational Linguistics",
"volume": "43",
"issue": "1",
"pages": "125--179",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ivan Habernal and Iryna Gurevych. 2017. Argumentation mining in user-generated web discourse. Computational Linguistics, 43(1):125-179, April.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Using complex argumentative interactions to reconstruct the argumentative structure of large-scale debates",
"authors": [
{
"first": "John",
"middle": [],
"last": "Lawrence",
"suffix": ""
},
{
"first": "Chris",
"middle": [],
"last": "Reed",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 4th Workshop on Argument Mining",
"volume": "",
"issue": "",
"pages": "108--117",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John Lawrence and Chris Reed. 2017. Using complex argumentative interactions to reconstruct the argumentative structure of large-scale debates. In Proceedings of the 4th Workshop on Argument Mining, pages 108-117, Copenhagen, Denmark, September. Association for Computational Linguistics.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Espresso: Leveraging generic patterns for automatically harvesting semantic relations",
"authors": [
{
"first": "Patrick",
"middle": [],
"last": "Pantel",
"suffix": ""
},
{
"first": "Marco",
"middle": [],
"last": "Pennacchiotti",
"suffix": ""
}
],
"year": 2006,
"venue": "Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "113--120",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Patrick Pantel and Marco Pennacchiotti. 2006. Espresso: Leveraging generic patterns for automatically harvesting semantic relations. In Proceedings of the 21st International Conference on Computational Linguistics and 44th Annual Meeting of the Association for Computational Linguistics, pages 113-120, Sydney, Australia, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Joint prediction in MST-style discourse parsing for argumentation mining",
"authors": [
{
"first": "Andreas",
"middle": [],
"last": "Peldszus",
"suffix": ""
},
{
"first": "Manfred",
"middle": [],
"last": "Stede",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "938--948",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andreas Peldszus and Manfred Stede. 2015. Joint prediction in MST-style discourse parsing for argumentation mining. In Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing, pages 938-948, Lisbon, Portugal, September. Association for Computational Linguistics.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Annotating argument components and relations in persuasive essays",
"authors": [
{
"first": "Christian",
"middle": [],
"last": "Stab",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers",
"volume": "",
"issue": "",
"pages": "1501--1510",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christian Stab and Iryna Gurevych. 2014. Annotating argument components and relations in persuasive essays. In Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1501-1510. Dublin City University and Association for Computational Linguistics.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Automatic argument quality assessment -New datasets and methods",
"authors": [
{
"first": "Assaf",
"middle": [],
"last": "Toledo",
"suffix": ""
},
{
"first": "Shai",
"middle": [],
"last": "Gretz",
"suffix": ""
},
{
"first": "Edo",
"middle": [],
"last": "Cohen-Karlik",
"suffix": ""
},
{
"first": "Roni",
"middle": [],
"last": "Friedman",
"suffix": ""
},
{
"first": "Elad",
"middle": [],
"last": "Venezian",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Lahav",
"suffix": ""
},
{
"first": "Michal",
"middle": [],
"last": "Jacovi",
"suffix": ""
},
{
"first": "Ranit",
"middle": [],
"last": "Aharonov",
"suffix": ""
},
{
"first": "Noam",
"middle": [],
"last": "Slonim",
"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": "5625--5635",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Assaf Toledo, Shai Gretz, Edo Cohen-Karlik, Roni Friedman, Elad Venezian, Dan Lahav, Michal Jacovi, Ranit Aharonov, and Noam Slonim. 2019. Automatic argument quality assessment -New datasets and methods. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th Inter- national Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 5625-5635. Association for Computational Linguistics.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Common round: Application of language technologies to large-scale web debates",
"authors": [
{
"first": "Hans",
"middle": [],
"last": "Uszkoreit",
"suffix": ""
},
{
"first": "Aleksandra",
"middle": [],
"last": "Gabryszak",
"suffix": ""
},
{
"first": "Leonhard",
"middle": [],
"last": "Hennig",
"suffix": ""
},
{
"first": "J\u00f6rg",
"middle": [],
"last": "Steffen",
"suffix": ""
},
{
"first": "Renlong",
"middle": [],
"last": "Ai",
"suffix": ""
},
{
"first": "Stephan",
"middle": [],
"last": "Busemann",
"suffix": ""
},
{
"first": "Jon",
"middle": [],
"last": "Dehdari",
"suffix": ""
},
{
"first": "Josef",
"middle": [],
"last": "Van Genabith",
"suffix": ""
},
{
"first": "Georg",
"middle": [],
"last": "Heigold",
"suffix": ""
},
{
"first": "Nils",
"middle": [],
"last": "Rethmeier",
"suffix": ""
},
{
"first": "Raphael",
"middle": [],
"last": "Rubino",
"suffix": ""
},
{
"first": "Sven",
"middle": [],
"last": "Schmeier",
"suffix": ""
},
{
"first": "Philippe",
"middle": [],
"last": "Thomas",
"suffix": ""
},
{
"first": "He",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Feiyu",
"middle": [],
"last": "Xu",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the Software Demonstrations of the 15th Conference of the European Chapter of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "5--8",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Hans Uszkoreit, Aleksandra Gabryszak, Leonhard Hennig, J\u00f6rg Steffen, Renlong Ai, Stephan Busemann, Jon De- hdari, Josef van Genabith, Georg Heigold, Nils Rethmeier, Raphael Rubino, Sven Schmeier, Philippe Thomas, He Wang, and Feiyu Xu. 2017. Common round: Application of language technologies to large-scale web debates. In Proceedings of the Software Demonstrations of the 15th Conference of the European Chapter of the Association for Computational Linguistics, pages 5-8, Valencia, Spain, April. Association for Computational Linguistics.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Computational argumentation quality assessment in natural language",
"authors": [
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Nona",
"middle": [],
"last": "Naderi",
"suffix": ""
},
{
"first": "Yufang",
"middle": [],
"last": "Hou",
"suffix": ""
},
{
"first": "Yonatan",
"middle": [],
"last": "Bilu",
"suffix": ""
},
{
"first": "Vinodkumar",
"middle": [],
"last": "Prabhakaran",
"suffix": ""
},
{
"first": "Tim",
"middle": [
"Alberdingk"
],
"last": "Thijm",
"suffix": ""
},
{
"first": "Graeme",
"middle": [],
"last": "Hirst",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics",
"volume": "1",
"issue": "",
"pages": "176--187",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Henning Wachsmuth, Nona Naderi, Yufang Hou, Yonatan Bilu, Vinodkumar Prabhakaran, Tim Alberdingk Thijm, Graeme Hirst, and Benno Stein. 2017a. Computational argumentation quality assessment in natural language. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguis- tics: Volume 1, Long Papers, pages 176-187. Association for Computational Linguistics.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Building an argument search engine for the web",
"authors": [
{
"first": "Henning",
"middle": [],
"last": "Wachsmuth",
"suffix": ""
},
{
"first": "Martin",
"middle": [],
"last": "Potthast",
"suffix": ""
},
{
"first": "Khalid",
"middle": [],
"last": "Al-Khatib",
"suffix": ""
},
{
"first": "Yamen",
"middle": [],
"last": "Ajjour",
"suffix": ""
},
{
"first": "Jana",
"middle": [],
"last": "Puschmann",
"suffix": ""
},
{
"first": "Jiani",
"middle": [],
"last": "Qu",
"suffix": ""
},
{
"first": "Jonas",
"middle": [],
"last": "Dorsch",
"suffix": ""
},
{
"first": "Viorel",
"middle": [],
"last": "Morari",
"suffix": ""
},
{
"first": "Janek",
"middle": [],
"last": "Bevendorff",
"suffix": ""
},
{
"first": "Benno",
"middle": [],
"last": "Stein",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 4th Workshop on Argument Mining",
"volume": "",
"issue": "",
"pages": "49--59",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Henning Wachsmuth, Martin Potthast, Khalid Al-Khatib, Yamen Ajjour, Jana Puschmann, Jiani Qu, Jonas Dorsch, Viorel Morari, Janek Bevendorff, and Benno Stein. 2017b. Building an argument search engine for the web. In Proceedings of the 4th Workshop on Argument Mining, pages 49-59. Association for Computational Linguistics.",
"links": null
},
"BIBREF22": {
"ref_id": "b22",
"title": "A corpus for research on deliberation and debate",
"authors": [
{
"first": "Marilyn",
"middle": [],
"last": "Walker",
"suffix": ""
},
{
"first": "Jean",
"middle": [
"Fox"
],
"last": "Tree",
"suffix": ""
},
{
"first": "Pranav",
"middle": [],
"last": "Anand",
"suffix": ""
},
{
"first": "Rob",
"middle": [],
"last": "Abbott",
"suffix": ""
},
{
"first": "Joseph",
"middle": [],
"last": "King",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)",
"volume": "",
"issue": "",
"pages": "812--817",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marilyn Walker, Jean Fox Tree, Pranav Anand, Rob Abbott, and Joseph King. 2012. A corpus for research on deliberation and debate. In Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12), pages 812-817, Istanbul, Turkey, May. European Language Resources Association (ELRA).",
"links": null
}
},
"ref_entries": {
"FIGREF1": {
"num": null,
"text": "Conceptual process of our semi-supervised bootstrapping approach: (a) Seed (ir)relevance patterns are selected manually from intially mined candidates. (b) New (ir)relevance patterns are mined and filtered automatically from text units matching the existing patterns, until no new patterns are found or k iterations have passed. (c) The corpus is cleaned by removing units matching the irrelevance patterns.",
"type_str": "figure",
"uris": null
},
"FIGREF2": {
"num": null,
"text": "(a) Histograms of the number of texts in the args.me corpus with a certain a number of irrelevant sentences, as detected by our approach. (b) Histogram of the number of detected (upper number) and removed (lower number) irrelevant sentences over the different sentence positions of a text.",
"type_str": "figure",
"uris": null
},
"TABREF0": {
"content": "<table><tr><td>Irrelevant #1</td></tr><tr><td>Relevant #2 (non-argumentat.)</td></tr><tr><td>Relevant #3-5 (argumentative)</td></tr><tr><td>Relevant #6-8 (argumentative)</td></tr><tr><td>Relevant #9-13 (argumentative)</td></tr><tr><td>Irrelevant #14</td></tr></table>",
"num": null,
"text": "Irrelevant #15https://www.debate.org/debates/Gay-Marriage/75/",
"html": null,
"type_str": "table"
},
"TABREF3": {
"content": "<table/>",
"num": null,
"text": "",
"html": null,
"type_str": "table"
}
}
}
} |