File size: 62,411 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 |
{
"paper_id": "2020",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T02:09:57.998697Z"
},
"title": "Annotation and Detection of Arguments in Tweets",
"authors": [
{
"first": "Robin",
"middle": [],
"last": "Schaefer",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Applied Computational Linguistics University of Potsdam",
"location": {
"settlement": "Potsdam",
"country": "Germany"
}
},
"email": "robin.schaefer@uni-potsdam.de"
},
{
"first": "Manfred",
"middle": [],
"last": "Stede",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Computational Linguistics University of Potsdam",
"location": {
"settlement": "Potsdam",
"country": "Germany"
}
},
"email": "stede@uni-potsdam.de"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Notwithstanding the increasing role Twitter plays in modern political and social discourse, resources built for conducting argument mining on tweets remain limited. In this paper, we present a new corpus of German tweets annotated for argument components. To the best of our knowledge, this is the first corpus containing not only annotated full tweets but also argumentative spans within tweets. We further report first promising results using supervised classification (F1: 0.82) and sequence labeling (F1: 0.72) approaches. 2 Related Work Related work on tweet-based argument mining has focused on separating argumentative tweets from non-argumentative ones and on defining new Twitter-specific tasks.",
"pdf_parse": {
"paper_id": "2020",
"_pdf_hash": "",
"abstract": [
{
"text": "Notwithstanding the increasing role Twitter plays in modern political and social discourse, resources built for conducting argument mining on tweets remain limited. In this paper, we present a new corpus of German tweets annotated for argument components. To the best of our knowledge, this is the first corpus containing not only annotated full tweets but also argumentative spans within tweets. We further report first promising results using supervised classification (F1: 0.82) and sequence labeling (F1: 0.72) approaches. 2 Related Work Related work on tweet-based argument mining has focused on separating argumentative tweets from non-argumentative ones and on defining new Twitter-specific tasks.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "In recent years the field of argument mining, which focuses on the automatic identification of argument components and their relations in text, has developed substantially (Stede and Schneider, 2018) . However, while the majority of research concentrates on well-structured documents (Moens et al., 2007; Stab and Gurevych, 2014) , less work has been done on user-generated web content (Park and Cardie, 2014; Habernal and Gurevych, 2015) . This shortcoming poses a problem as systems trained on formal and edited texts tend to be inapt of extracting patterns from the more informal user-generated content (\u0160najder, 2016) .",
"cite_spans": [
{
"start": 172,
"end": 199,
"text": "(Stede and Schneider, 2018)",
"ref_id": null
},
{
"start": 284,
"end": 304,
"text": "(Moens et al., 2007;",
"ref_id": "BIBREF14"
},
{
"start": 305,
"end": 329,
"text": "Stab and Gurevych, 2014)",
"ref_id": "BIBREF18"
},
{
"start": 386,
"end": 409,
"text": "(Park and Cardie, 2014;",
"ref_id": "BIBREF15"
},
{
"start": 410,
"end": 438,
"text": "Habernal and Gurevych, 2015)",
"ref_id": "BIBREF11"
},
{
"start": 606,
"end": 621,
"text": "(\u0160najder, 2016)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "In this paper we focus on tweets, which are of great interest for the argument mining community due to the increasing use of the microblogging service Twitter 1 in political online discourse. While some first work on argument mining in tweets exists (Addawood and Bashir, 2016; Dusmanu et al., 2017) , only a small number of available annotated corpora have been created that can be utilized for training tweet-specific argument mining systems (Bosc et al., 2016) .",
"cite_spans": [
{
"start": 250,
"end": 277,
"text": "(Addawood and Bashir, 2016;",
"ref_id": "BIBREF0"
},
{
"start": 278,
"end": 299,
"text": "Dusmanu et al., 2017)",
"ref_id": "BIBREF7"
},
{
"start": 444,
"end": 463,
"text": "(Bosc et al., 2016)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "To improve on this point, we present a new corpus of German tweets annotated for claim and evidence 2 . To the best of our knowledge, this is the first argument tweet corpus not exclusively annotated with the full tweet as the unit of annotation. Instead, argumentative spans within tweets, henceforth called argumentative discourse units (ADU) (Peldszus and Stede, 2013) , have been annotated as well. They render the corpus suitable not only for supervised classification but also for sequence labeling approaches. We also present first promising experimental results using this corpus.",
"cite_spans": [
{
"start": 345,
"end": 371,
"text": "(Peldszus and Stede, 2013)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This paper is structured as follows: Section 2 gives a short overview of the relevant social media and Twitter-related literature on argument mining. Section 3 describes the corpus, the annotation scheme and the annotation procedure. In Section 4 we present first classification and sequence labeling results using the annotated data. Section 5 discusses our results and gives a brief outlook. Addawood and Bashir (2016) present a corpus of English tweets annotated for arguments and evidence types like news media accounts or expert opinions. First, arguments are identified on the full tweet level, followed by the subsequent annotation of evidence types. Annotators achieved Cohen's Kappa scores of 0.67 and 0.79, respectively. An SVM trained on linguistic and Twitter-related features yielded an F1 score of 0.89 on the binary classification task (non-argumentative vs argumentative). Bosc et al. (2016) describe DART, a Twitter argument corpus annotated for arguments and their relations. In contrast to our work, they do not distinguish claim from evidence but join them in the category argumentative. Again, annotations are conducted on the full tweet level and result in a Krippendorff alpha score of 0.81. This corpus is used by Dusmanu et al. (2017) for argument classification. Using a set of lexical, Twitter-specific, semantic and sentiment features, they achieved an F1 score of 0.78 on the binary classification task (non-argumentative vs argumentative). They further investigated approaches to perform fact recognition and source identification. Wojatzki and Zesch (2016) propose an alternative approach to argument mining in tweets. Specifically, they reconsider the challenging problem of implicit claim detection as a stance classification problem by reformulating implicit claims as implicit stances. This procedure is based on the assumption that an implicit stance can be more easily inferred from the respective tweet. They present the Atheism Stance Corpus, which contains tweets annotated for implicit stances. An SVM trained on token and character n-grams yielded an F1 score of 0.66. Schaefer and Stede (2019) improve on these results using different word and sentence embeddings (F1: 0.78). Goudas et al. (2014) offer early results for argument mining not specifically on Twitter but on social media. They apply classification to separate non-argumentative from argumentative texts. In a subsequent step, sequence labeling is used to extract ADUs from the latter. This two-step approach makes their work comparable to ours. They report F1 scores of 0.77 and 0.42 for the two tasks, respectively.",
"cite_spans": [
{
"start": 394,
"end": 420,
"text": "Addawood and Bashir (2016)",
"ref_id": "BIBREF0"
},
{
"start": 889,
"end": 907,
"text": "Bosc et al. (2016)",
"ref_id": "BIBREF4"
},
{
"start": 1238,
"end": 1259,
"text": "Dusmanu et al. (2017)",
"ref_id": "BIBREF7"
},
{
"start": 1562,
"end": 1587,
"text": "Wojatzki and Zesch (2016)",
"ref_id": "BIBREF20"
},
{
"start": 2219,
"end": 2239,
"text": "Goudas et al. (2014)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Our complete initial corpus consists of 77,100 tweets collected in 2019 via the Twitter API using the Python library Tweepy 3 . All tweets contain the keyword klima (\"climate\") and mainly concentrate on the topic of climate change, which was intensely discussed by German media and politics during that time. We conducted the following preprocessing steps.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Corpus Annotation",
"sec_num": "3"
},
{
"text": "First, we removed all retweets and excluded non-German tweets using the language identification tool langid (Lui and Baldwin, 2012) . These steps led to a subset of 29,525 tweets. In the following, we grouped the tweets into pairs, consisting of a tweet, henceforth called context tweet, and the tweet to be annotated, which is a reply to the context tweet and, for this reason, is called reply tweet. This approach is motivated by the assumption that tweets in a reply relation are more likely to contain argumentation (Dykes et al., 2020) . Moreover, given the short nature of tweets, providing a context is supposed to help interpreting the reply tweet's content. All tweets that were no replies were removed and missing context tweets were collected in an additional step. Finally, we removed all @-mentions at the beginning of a tweet, as these mainly point to the tweet's recipients. The final corpus consists of 12,296 context and reply tweet pairs. For the present study, a subset of 300 tweet pairs was annotated. 4",
"cite_spans": [
{
"start": 108,
"end": 131,
"text": "(Lui and Baldwin, 2012)",
"ref_id": null
},
{
"start": 520,
"end": 540,
"text": "(Dykes et al., 2020)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Corpus Annotation",
"sec_num": "3"
},
{
"text": "We focus on the two main components of argumentation: claim and evidence. We define a claim as a standpoint towards the topic being discussed (i.e. climate change). In contrast, an evidence unit is a statement used to support or attack such a standpoint. Hence, the crucial difference between claim and evidence is the characteristic of evidence units being always related to another statement while claims can be independent units. We distinguish further between evidence 1) relating to a claim in the reply tweet, 2) relating to a claim in the context tweet and 3) relating to claims in both tweets. Importantly, we do not define an ADU syntactically, e.g. by focusing exclusively on the clause or sentence level.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Scheme",
"sec_num": "3.1"
},
{
"text": "Due to the informal language used in the tweets we consider it appropriate to allow the annotators some flexibility to decide on the actual ADU span.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Scheme",
"sec_num": "3.1"
},
{
"text": "As distinguishing between claim and evidence can be a quite subjective task, especially on Twitter, annotators were advised to follow our component definitions as close as possible. Statements that function independently of other statements shall be annotated as claims. However, if a statement refers to another proposition either by supporting or attacking it or by giving additional information it shall be annotated as evidence, despite its potential usability as a claim. Therefore, annotators were further instructed to focus on possible causal relationships (in a wide sense) between two statements. If a statement directly follows from another it is likely to be a claim (e.g. [We have to limit CO2 emissions] claim , [as too much CO2 has been shown to increase the greenhouse effect.] evidence ). We found that using this strategy to decide on the direction of the argumentation, i.e. which ADU is evidence and which ADU is the claim, facilitated the annotation procedure notably. For our purposes, we do not differentiate between correct and incorrect statements. We also do not explicitly annotate relations between two components.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Scheme",
"sec_num": "3.1"
},
{
"text": "Two annotators, one of which is a co-author of this paper, were trained in an iterative two-step procedure. First, both annotators individually labelled a subset of 20 tweet pairs according to the annotation scheme. They compared their results, discussed different interpretations and tried to consolidate them. This procedure was repeated until both annotators felt comfortable in completing the task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Procedure and Results",
"sec_num": "3.2"
},
{
"text": "For the actual annotation study we again used a two-step approach. Annotators first had to answer two multiple choice questions asking if a claim or evidence can be identified in the reply tweet. Only if one of the two components was found the annotator would continue to the ADU annotation step. No restrictions on the allowed maximal number of components per tweet were made, as this could potentially have led to differing choices in longer tweets. While annotations themselves only were created for ADU spans, we also derived separate tweet-level annotation sets for claim, evidence and argument (claim or evidence) annotations. Also, we experimented with analysing annotations both on the tweet and the ADU level.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Procedure and Results",
"sec_num": "3.2"
},
{
"text": "First, we present mean percentages of the ADU annotation frequencies. Of the 300 tweets 14% were annotated as non-argumentative. 27% of the tweets contained exactly one ADU (25%: claim; 2%: evidence). 59% of the tweets were annotated for multiple ADUs (27%: 1 claim & 1 evidence unit; 2% 1 claim & >1 evidence units; 15%: >1 claims & 1 evidence unit) which demonstrates the need for ADU-level annotation even in short texts like tweets.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Annotation Procedure and Results",
"sec_num": "3.2"
},
{
"text": "Claim Evidence Cohen's Kappa 0.55 0.37 We calculated Cohen's Kappa scores to measure Inter Annotator Agreement (IAA) (Artstein and Poesio, 2008) . As shown in Table 1 , results for the claim and evidence questions were 0.55 and 0.37, respectively, which indicates that deciding on the presence of evidence is more subjective. This pattern returns in the scores based on the annotations on the tweet level (Table 2) . Whereas results for argument and claim annotations are somewhat similar, the kappa for evidence annotation is reduced. Further, the results show that the multi class annotation (claim vs evidence vs non-argumentative) is particularly difficult. As this task is somewhat subjective in nature, a drop of performance is expected. Although we are aware that the IAA results are relatively low, we consider them acceptable due to the subtlety of the task. This is in line with the interpretation of annotation results by Aharoni et al. (2014) , who report 0.39 and 0.4 for claim and evidence annotation tasks, respectively.",
"cite_spans": [
{
"start": 117,
"end": 144,
"text": "(Artstein and Poesio, 2008)",
"ref_id": "BIBREF3"
},
{
"start": 933,
"end": 954,
"text": "Aharoni et al. (2014)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [
{
"start": 159,
"end": 166,
"text": "Table 1",
"ref_id": "TABREF0"
},
{
"start": 405,
"end": 414,
"text": "(Table 2)",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Metric",
"sec_num": null
},
{
"text": "In this section, we present first experimental results based on the annotated corpus. We apply two different approaches: For the tweet-level annotations we trained supervised classification models. This is comparable to the prior studies of Addawood and Bashir (2016) and Dusmanu et al. (2017) . In addition, we use the ADU-level annotations for running a sequence labeling approach similar to Goudas et al. (2014) . We experimented with different combinations of feature sets, preprocessing steps and models. However, we only present the best results here. Table 3 : Classification Results (l = lowercase, p = punctuation, s = stopword, w = weighted)",
"cite_spans": [
{
"start": 241,
"end": 267,
"text": "Addawood and Bashir (2016)",
"ref_id": "BIBREF0"
},
{
"start": 272,
"end": 293,
"text": "Dusmanu et al. (2017)",
"ref_id": "BIBREF7"
},
{
"start": 394,
"end": 414,
"text": "Goudas et al. (2014)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [
{
"start": 558,
"end": 565,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experiments and Results",
"sec_num": "4"
},
{
"text": "Tweet level. Classification models were trained on different combinations of n-grams and on pretrained BERT-based document embeddings (Devlin et al., 2019) . The latter were created using FLAIR, an NLP framework that contains a unified interface for employing different types of text embeddings (Akbik et al., 2019) . All shown classification results are yielded using eXtreme Gradient Boosting (XGBoost) (Chen and Guestrin, 2016) , which is a variant of the Gradient Boosting approach introduced by Friedman (2000). We implemented three different classification tasks based on the respective binary target sets: argumentative vs non-argumentative, claim vs no claim or evidence vs no evidence. All results are 10-fold cross-validated. Table 3 shows macro F1, precision and recall scores, which are weighted for the unbalanced distribution of classes. Pretrained BERT embeddings yield better F1 scores for argument (0.82 vs 0.8) and claim (0.82 vs 0.79) classifications. Interestingly, a model trained on uni-and bigrams performs better on the evidence task than the BERT-based model (0.67 vs 0.59). Importantly, scores for the argument and claim tasks are substantially higher than for the evidence task. ADU level. Sequence labeling models were trained on the following features: 1) unigrams, 2) a combination of linguistic (e.g., n-grams, POS Tags) and Twitter-related (e.g., hashtags, @-mentions) features, 3) pretrained BERT-based word embeddings, which were again created using FLAIR. We chose a Conditional Random Fields approach (Lafferty et al., 2001 ), using the sklearn-crfsuite 5 . Again, all results are from 10-fold cross-validation. In the sequence labeling approach BERT-based models perform best for all three labeling tasks. Using a set of linguistic and Twitter-related features improves the F1 scores compared to the simple unigram models in the argument (0.7 vs 0.69) and claim (0.56 vs 0.53) tasks. However, no improvement is achieved in the evidence task. Interestingly, scores are highest for the evidence task whereas the results for the claim task are considerably lower. This pattern contrasts with the tweet-level classification results.",
"cite_spans": [
{
"start": 134,
"end": 155,
"text": "(Devlin et al., 2019)",
"ref_id": "BIBREF6"
},
{
"start": 295,
"end": 315,
"text": "(Akbik et al., 2019)",
"ref_id": "BIBREF2"
},
{
"start": 405,
"end": 430,
"text": "(Chen and Guestrin, 2016)",
"ref_id": "BIBREF5"
},
{
"start": 1537,
"end": 1559,
"text": "(Lafferty et al., 2001",
"ref_id": "BIBREF12"
}
],
"ref_spans": [
{
"start": 736,
"end": 743,
"text": "Table 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Features",
"sec_num": null
},
{
"text": "In this paper we presented a new corpus of German tweets annotated for claim and evidence. While a few previous studies on tweet corpus creation for argument mining exist (Bosc et al., 2016) , to the best of our knowledge our corpus is the first tweet dataset with ADU annotations. It is also the first German tweet dataset generally annotated for argumentation.",
"cite_spans": [
{
"start": 171,
"end": 190,
"text": "(Bosc et al., 2016)",
"ref_id": "BIBREF4"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion and Outlook",
"sec_num": "5"
},
{
"text": "Although we showed that due to the subtlety of the task relatively low IAA scores were achieved, classification and sequence labeling results based on the dataset are promising. Classifying argument and claim components led to robust F1 scores around 0.8. Solely evidence units posed somewhat of a challenge for the classifier. However, sequence labeling models performed best for evidence units. With both approaches we surpassed the results presented by Goudas et al. (2014) .",
"cite_spans": [
{
"start": 456,
"end": 476,
"text": "Goudas et al. (2014)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion and Outlook",
"sec_num": "5"
},
{
"text": "Given that the IAA scores for evidence annotations were reduced as well, we conclude that evidence units pose an especially hard problem to solve. Recalling our definitions of claim and evidence, this seems intuitive. As evidence units are only defined with respect to claims, a decision has to be made about the exact boundary between both components. Moreover, since tweets tend to contain a high degree of implicitness, it can be demanding to judge if a sequence in fact is relating to a claim. We plan to take this issue into account by refining our annotation scheme further.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion and Outlook",
"sec_num": "5"
},
{
"text": "Another interesting path of future work will be the continuing development of the argument detector. Following Goudas et al. (2014) , one possible way of enhancing results could be building a pipeline based on both classification and sequence labeling approaches. More specifically, a classifier customized for identifying argumentative tweets could function as a filter, thereby allowing to train a sequence labeling model on a purely argumentative tweet set. This could increase the model's precision. To this end, we intend to enlarge the number of annotated data.",
"cite_spans": [
{
"start": 111,
"end": 131,
"text": "Goudas et al. (2014)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discussion and Outlook",
"sec_num": "5"
},
{
"text": "https://www.tweepy.org/ 4 Corpus repository: https://github.com/RobinSchaefer/climate-tweet-corpus.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "sklearn-crfsuite (https://sklearn-crfsuite.readthedocs.io) is a scikit-learn wrapper based on CRFsuite (http://www.chokkan.org/software/crfsuite/).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "We would like to thank Polina Krasilnikova for assisting in annotating our data and Crowdee (https: //www.crowdee.com/) for support with their annotation environment. We further thank the anonymous reviewers for their helpful comments.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "what is your evidence?\" a study of controversial topics on social media",
"authors": [
{
"first": "Aseel",
"middle": [],
"last": "Addawood",
"suffix": ""
},
{
"first": "Masooda",
"middle": [],
"last": "Bashir",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Third Workshop on Argument Mining (ArgMining2016)",
"volume": "",
"issue": "",
"pages": "1--11",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Aseel Addawood and Masooda Bashir. 2016. \"what is your evidence?\" a study of controversial topics on social media. In Proceedings of the Third Workshop on Argument Mining (ArgMining2016), pages 1-11, Berlin, Germany, August. Association for Computational Linguistics.",
"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": "FLAIR: An easy-to-use framework for state-of-the-art NLP",
"authors": [
{
"first": "Alan",
"middle": [],
"last": "Akbik",
"suffix": ""
},
{
"first": "Tanja",
"middle": [],
"last": "Bergmann",
"suffix": ""
},
{
"first": "Duncan",
"middle": [],
"last": "Blythe",
"suffix": ""
},
{
"first": "Kashif",
"middle": [],
"last": "Rasul",
"suffix": ""
},
{
"first": "Stefan",
"middle": [],
"last": "Schweter",
"suffix": ""
},
{
"first": "Roland",
"middle": [],
"last": "Vollgraf",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics (Demonstrations)",
"volume": "",
"issue": "",
"pages": "54--59",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alan Akbik, Tanja Bergmann, Duncan Blythe, Kashif Rasul, Stefan Schweter, and Roland Vollgraf. 2019. FLAIR: An easy-to-use framework for state-of-the-art NLP. In Proceedings of the 2019 Conference of the North Amer- ican Chapter of the Association for Computational Linguistics (Demonstrations), pages 54-59, Minneapolis, Minnesota, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Inter-coder agreement for computational linguistics",
"authors": [
{
"first": "Ron",
"middle": [],
"last": "Artstein",
"suffix": ""
},
{
"first": "Massimo",
"middle": [],
"last": "Poesio",
"suffix": ""
}
],
"year": 2008,
"venue": "Comput. Linguist",
"volume": "34",
"issue": "4",
"pages": "555--596",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ron Artstein and Massimo Poesio. 2008. Inter-coder agreement for computational linguistics. Comput. Linguist., 34(4):555-596, December.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "DART: a dataset of arguments and their relations on twitter",
"authors": [
{
"first": "Tom",
"middle": [],
"last": "Bosc",
"suffix": ""
},
{
"first": "Elena",
"middle": [],
"last": "Cabrio",
"suffix": ""
},
{
"first": "Serena",
"middle": [],
"last": "Villata",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16)",
"volume": "",
"issue": "",
"pages": "1258--1263",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tom Bosc, Elena Cabrio, and Serena Villata. 2016. DART: a dataset of arguments and their relations on twitter. In Proceedings of the Tenth International Conference on Language Resources and Evaluation (LREC'16), pages 1258-1263, Portoro\u017e, Slovenia, May. European Language Resources Association (ELRA).",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "XGBoost: A scalable tree boosting system",
"authors": [
{
"first": "Tianqi",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Carlos",
"middle": [],
"last": "Guestrin",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD '16",
"volume": "",
"issue": "",
"pages": "785--794",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Tianqi Chen and Carlos Guestrin. 2016. XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, KDD '16, pages 785-794, New York, NY, USA. Association for Computing Machinery.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "BERT: Pre-training of deep bidirectional transformers for language understanding",
"authors": [
{
"first": "Jacob",
"middle": [],
"last": "Devlin",
"suffix": ""
},
{
"first": "Ming-Wei",
"middle": [],
"last": "Chang",
"suffix": ""
},
{
"first": "Kenton",
"middle": [],
"last": "Lee",
"suffix": ""
},
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
"volume": "1",
"issue": "",
"pages": "4171--4186",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirec- tional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers), pages 4171-4186, Minneapolis, Minnesota, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Argument mining on twitter: Arguments, facts and sources",
"authors": [
{
"first": "Mihai",
"middle": [],
"last": "Dusmanu",
"suffix": ""
},
{
"first": "Elena",
"middle": [],
"last": "Cabrio",
"suffix": ""
},
{
"first": "Serena",
"middle": [],
"last": "Villata",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "2317--2322",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mihai Dusmanu, Elena Cabrio, and Serena Villata. 2017. Argument mining on twitter: Arguments, facts and sources. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 2317-2322, Copenhagen, Denmark, September. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Reconstructing arguments from noisy text",
"authors": [
{
"first": "Natalie",
"middle": [],
"last": "Dykes",
"suffix": ""
},
{
"first": "Stefan",
"middle": [],
"last": "Evert",
"suffix": ""
},
{
"first": "Merlin",
"middle": [],
"last": "G\u00f6ttlinger",
"suffix": ""
},
{
"first": "Philipp",
"middle": [],
"last": "Heinrich",
"suffix": ""
},
{
"first": "Lutz",
"middle": [],
"last": "Schr\u00f6der",
"suffix": ""
}
],
"year": 2020,
"venue": "Datenbank-Spektrum",
"volume": "20",
"issue": "2",
"pages": "123--129",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Natalie Dykes, Stefan Evert, Merlin G\u00f6ttlinger, Philipp Heinrich, and Lutz Schr\u00f6der. 2020. Reconstructing arguments from noisy text. Datenbank-Spektrum, 20(2):123-129.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Greedy function approximation: A gradient boosting machine",
"authors": [
{
"first": "H",
"middle": [],
"last": "Jerome",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Friedman",
"suffix": ""
}
],
"year": 2000,
"venue": "Annals of Statistics",
"volume": "29",
"issue": "",
"pages": "1189--1232",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jerome H. Friedman. 2000. Greedy function approximation: A gradient boosting machine. Annals of Statistics, 29:1189-1232.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Argument extraction from news, blogs, and social media",
"authors": [
{
"first": "Theodosis",
"middle": [],
"last": "Goudas",
"suffix": ""
},
{
"first": "Christos",
"middle": [],
"last": "Louizos",
"suffix": ""
}
],
"year": 2014,
"venue": "Aristidis Likas, Konstantinos Blekas, and Dimitris Kalles",
"volume": "",
"issue": "",
"pages": "287--299",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Theodosis Goudas, Christos Louizos, Georgios Petasis, and Vangelis Karkaletsis. 2014. Argument extraction from news, blogs, and social media. In Aristidis Likas, Konstantinos Blekas, and Dimitris Kalles, editors, Artificial Intelligence: Methods and Applications, pages 287-299, Cham. Springer International Publishing.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Exploiting debate portals for semi-supervised argumentation mining in user-generated web discourse",
"authors": [
{
"first": "Ivan",
"middle": [],
"last": "Habernal",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2015,
"venue": "Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "2127--2137",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ivan Habernal and Iryna Gurevych. 2015. Exploiting debate portals for semi-supervised argumentation mining in user-generated web discourse. In Proceedings of the 2015 Conference on Empirical Methods in Natural Lan- guage Processing, pages 2127-2137, Lisbon, Portugal, September. Association for Computational Linguistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Conditional random fields: Probabilistic models for segmenting and labeling sequence data",
"authors": [
{
"first": "John",
"middle": [
"D"
],
"last": "Lafferty",
"suffix": ""
},
{
"first": "Andrew",
"middle": [],
"last": "Mccallum",
"suffix": ""
},
{
"first": "Fernando",
"middle": [
"C N"
],
"last": "Pereira",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings of the Eighteenth International Conference on Machine Learning, ICML '01",
"volume": "",
"issue": "",
"pages": "282--289",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In Proceedings of the Eighteenth International Conference on Machine Learning, ICML '01, page 282-289, San Francisco, CA, USA. Morgan Kaufmann Publishers Inc.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "2012. langid.py: An off-the-shelf language identification tool",
"authors": [
{
"first": "Marco",
"middle": [],
"last": "Lui",
"suffix": ""
},
{
"first": "Timothy",
"middle": [],
"last": "Baldwin",
"suffix": ""
}
],
"year": null,
"venue": "Proceedings of the ACL 2012 System Demonstrations",
"volume": "",
"issue": "",
"pages": "25--30",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marco Lui and Timothy Baldwin. 2012. langid.py: An off-the-shelf language identification tool. In Proceedings of the ACL 2012 System Demonstrations, pages 25-30, Jeju Island, Korea, July. Association for Computational Linguistics.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Automatic detection of arguments in legal texts",
"authors": [
{
"first": "Marie-Francine",
"middle": [],
"last": "Moens",
"suffix": ""
},
{
"first": "Erik",
"middle": [],
"last": "Boiy",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the 11th International Conference on Artificial Intelligence and Law, ICAIL '07",
"volume": "",
"issue": "",
"pages": "225--230",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Marie-Francine Moens, Erik Boiy, Raquel Mochales Palau, and Chris Reed. 2007. Automatic detection of argu- ments in legal texts. In Proceedings of the 11th International Conference on Artificial Intelligence and Law, ICAIL '07, page 225-230, New York, NY, USA. Association for Computing Machinery.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Identifying appropriate support for propositions in online user comments",
"authors": [
{
"first": "Joonsuk",
"middle": [],
"last": "Park",
"suffix": ""
},
{
"first": "Claire",
"middle": [],
"last": "Cardie",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the First Workshop on Argumentation Mining",
"volume": "",
"issue": "",
"pages": "29--38",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Joonsuk Park and Claire Cardie. 2014. Identifying appropriate support for propositions in online user comments. In Proceedings of the First Workshop on Argumentation Mining, pages 29-38, Baltimore, Maryland, June. Association for Computational Linguistics.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "From argument diagrams to argumentation mining in texts: A survey",
"authors": [
{
"first": "Andreas",
"middle": [],
"last": "Peldszus",
"suffix": ""
},
{
"first": "Manfred",
"middle": [],
"last": "Stede",
"suffix": ""
}
],
"year": 2013,
"venue": "Int. J. Cogn. Inform. Nat. Intell",
"volume": "7",
"issue": "1",
"pages": "1--31",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Andreas Peldszus and Manfred Stede. 2013. From argument diagrams to argumentation mining in texts: A survey. Int. J. Cogn. Inform. Nat. Intell., 7(1):1-31, January.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Improving implicit stance classification in tweets using word and sentence embeddings",
"authors": [
{
"first": "Robin",
"middle": [],
"last": "Schaefer",
"suffix": ""
},
{
"first": "Manfred",
"middle": [],
"last": "Stede",
"suffix": ""
}
],
"year": 2019,
"venue": "KI 2019: Advances in Artificial Intelligence",
"volume": "",
"issue": "",
"pages": "299--307",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Robin Schaefer and Manfred Stede. 2019. Improving implicit stance classification in tweets using word and sentence embeddings. In Christoph Benzm\u00fcller and Heiner Stuckenschmidt, editors, KI 2019: Advances in Artificial Intelligence, pages 299-307, Cham. Springer International Publishing.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Identifying argumentative discourse structures in persuasive essays",
"authors": [
{
"first": "Christian",
"middle": [],
"last": "Stab",
"suffix": ""
},
{
"first": "Iryna",
"middle": [],
"last": "Gurevych",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP)",
"volume": "",
"issue": "",
"pages": "46--56",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christian Stab and Iryna Gurevych. 2014. Identifying argumentative discourse structures in persuasive essays. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 46-56, Doha, Qatar, October. Association for Computational Linguistics.",
"links": null
},
"BIBREF20": {
"ref_id": "b20",
"title": "Stance-based Argument Mining -Modeling Implicit Argumentation Using Stance",
"authors": [
{
"first": "Michael",
"middle": [],
"last": "Wojatzki",
"suffix": ""
},
{
"first": "Torsten",
"middle": [],
"last": "Zesch",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the KONVENS",
"volume": "",
"issue": "",
"pages": "313--322",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Michael Wojatzki and Torsten Zesch. 2016. Stance-based Argument Mining -Modeling Implicit Argumentation Using Stance. In Proceedings of the KONVENS, pages 313-322.",
"links": null
},
"BIBREF21": {
"ref_id": "b21",
"title": "Social media argumentation mining: The quest for deliberateness in raucousness",
"authors": [
{
"first": "",
"middle": [],
"last": "Jan\u0161najder",
"suffix": ""
}
],
"year": 2016,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jan\u0160najder. 2016. Social media argumentation mining: The quest for deliberateness in raucousness.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"content": "<table><tr><td>Metric</td><td colspan=\"5\">Multi (s) Argumentative (s) Argumentative (t) Claim (t) Evidence (t)</td></tr><tr><td>Cohen's Kappa</td><td>0.38</td><td>0.45</td><td>0.53</td><td>0.55</td><td>0.44</td></tr></table>",
"html": null,
"type_str": "table",
"num": null,
"text": "Inter Annotator Agreement (Questions)"
},
"TABREF1": {
"content": "<table/>",
"html": null,
"type_str": "table",
"num": null,
"text": "Inter Annotator Agreement (s = ADU span, t = full tweet)"
},
"TABREF4": {
"content": "<table/>",
"html": null,
"type_str": "table",
"num": null,
"text": "Sequence Labeling Results (w = weighted)"
}
}
}
} |