File size: 66,906 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 | {
"paper_id": "2022",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T05:58:15.899607Z"
},
"title": "Bi-Directional Recurrent Neural Ordinary Differential Equations for Social Media Text Classification",
"authors": [
{
"first": "Maunika",
"middle": [],
"last": "Tamire",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Indian Institute of Technology Hyderabad",
"location": {
"country": "India"
}
},
"email": ""
},
{
"first": "Srinivas",
"middle": [],
"last": "Anumasa",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Indian Institute of Technology Hyderabad",
"location": {
"country": "India"
}
},
"email": ""
},
{
"first": "P",
"middle": [
"K"
],
"last": "Srijith",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Indian Institute of Technology Hyderabad",
"location": {
"country": "India"
}
},
"email": "srijith@cse.iith.ac.in"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Classification of posts in social media such as Twitter is difficult due to the noisy and short nature of texts. Sequence classification models based on recurrent neural networks (RNN) are popular for classifying posts that are sequential in nature. RNNs assume the hidden representation dynamics to evolve in a discrete manner and do not consider the exact time of the posting. In this work, we propose to use recurrent neural ordinary differential equations (RN-ODE) for social media post classification which consider the time of posting and allow the computation of hidden representation to evolve in a time-sensitive continuous manner. In addition, we propose a novel model, Bi-directional RNODE (Bi-RNODE), which can consider the information flow in both the forward and backward directions of posting times to predict the post label. Our experiments demonstrate that RNODE and Bi-RNODE are effective for the problem of stance classification of rumours in social media.",
"pdf_parse": {
"paper_id": "2022",
"_pdf_hash": "",
"abstract": [
{
"text": "Classification of posts in social media such as Twitter is difficult due to the noisy and short nature of texts. Sequence classification models based on recurrent neural networks (RNN) are popular for classifying posts that are sequential in nature. RNNs assume the hidden representation dynamics to evolve in a discrete manner and do not consider the exact time of the posting. In this work, we propose to use recurrent neural ordinary differential equations (RN-ODE) for social media post classification which consider the time of posting and allow the computation of hidden representation to evolve in a time-sensitive continuous manner. In addition, we propose a novel model, Bi-directional RNODE (Bi-RNODE), which can consider the information flow in both the forward and backward directions of posting times to predict the post label. Our experiments demonstrate that RNODE and Bi-RNODE are effective for the problem of stance classification of rumours in social media.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Information disseminated in social media such as Twitter can be useful for addressing several realworld problems like rumour detection, disaster management, and opinion mining. Most of these problems involve classifying social media posts into different categories based on their textual content. For example, classifying the veracity of tweets as False, True, or unverified allows one to debunk the rumours evolving in social media (Zubiaga et al., 2018a) . However, social media text is extremely noisy with informal grammar, typographical errors, and irregular vocabulary. In addition, the character limit (240 characters) imposed by social media such as Twitter make it even harder to perform text classification.",
"cite_spans": [
{
"start": 433,
"end": 456,
"text": "(Zubiaga et al., 2018a)",
"ref_id": "BIBREF17"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Social media text classification, such as rumour stance classification 1 (Qazvinian et al., 1 Rumour stance classification helps to identify the veracity 2011; Zubiaga et al., 2016; Lukasik et al., 2019) can be addressed effectively using sequence labelling models such as long short term memory (LSTM) networks (Zubiaga et al., 2016; Augenstein et al., 2016; Kochkina et al., 2017; Zubiaga et al., 2018b,a; Dey et al., 2018; Liu et al., 2019; Tian et al., 2020) . Though they consider the sequential nature of tweets, they ignore the temporal aspects associated with the tweets. The time gap between tweets varies a lot and LSTMs ignore this irregularity in tweet occurrences. They are discrete state space models where hidden representation changes from one tweet to another without considering the time difference between the tweets. Considering the exact times at which tweets occur can play an important role in determining the label. If the time gap between tweets is large, then the corresponding labels may not influence each other but can have a very high influence if they are closer.",
"cite_spans": [
{
"start": 92,
"end": 93,
"text": "1",
"ref_id": null
},
{
"start": 160,
"end": 181,
"text": "Zubiaga et al., 2016;",
"ref_id": "BIBREF18"
},
{
"start": 182,
"end": 203,
"text": "Lukasik et al., 2019)",
"ref_id": "BIBREF11"
},
{
"start": 312,
"end": 334,
"text": "(Zubiaga et al., 2016;",
"ref_id": "BIBREF18"
},
{
"start": 335,
"end": 359,
"text": "Augenstein et al., 2016;",
"ref_id": "BIBREF0"
},
{
"start": 360,
"end": 382,
"text": "Kochkina et al., 2017;",
"ref_id": "BIBREF9"
},
{
"start": 383,
"end": 407,
"text": "Zubiaga et al., 2018b,a;",
"ref_id": null
},
{
"start": 408,
"end": 425,
"text": "Dey et al., 2018;",
"ref_id": "BIBREF4"
},
{
"start": 426,
"end": 443,
"text": "Liu et al., 2019;",
"ref_id": "BIBREF10"
},
{
"start": 444,
"end": 462,
"text": "Tian et al., 2020)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We propose to use recurrent neural ordinary differential equations (RNODE) (Rubanova et al., 2019) and developed a novel approach bidirectional RNODE (Bi-RNODE), which can naturally consider the temporal information to perform time sensitive classification of social media posts. NODE (Chen et al., 2018 ) is a continuous depth deep learning model that performs transformation of feature vectors in a continuous manner using ordinary differential equation solvers. NODEs bring parameter efficiency and address model selection in deep learning to a great extent. RNODE generalizes RNN by extending NODE for time-series data by considering temporal information associated with the sequential data. Hidden representations are changed continuously by considering the temporal information.",
"cite_spans": [
{
"start": 75,
"end": 98,
"text": "(Rubanova et al., 2019)",
"ref_id": "BIBREF13"
},
{
"start": 280,
"end": 303,
"text": "NODE (Chen et al., 2018",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We propose to use RNODE for the task of sequence labeling of posts, which considers arrival times of the posts for updating hidden representaof a rumour post by classifying the reply tweets into different stance classes such as Support, Deny, Question, Comment tions and for classifying the post. In addition, we propose a novel model, Bi-RNODE, which considers not only information from the past but also from the future in predicting the label of the post. Here, continuously evolving hidden representations in the forward and backward directions in time are combined and used to predict the post label. We show the effectiveness of the proposed models on the rumour stance classification problem in Twitter using the RumourEval-2019 (Derczynski et al., 2019) dataset. We found RNODE and Bi-RNODE can improve the social media text classification by effectively making use of the temporal information and is better than LSTMs and gated recurrent units (GRU) with temporal features.",
"cite_spans": [
{
"start": 736,
"end": 761,
"text": "(Derczynski et al., 2019)",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We consider the problem of classifying social media posts into different classes. Let D be a collec-",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "tion of N posts, D = {p i } N i=1",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": ". Each post p i is assumed to be a tuple containing information such as textual and contextual features x i , time of the post t i and the label associated with the post y i , thus",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "p i = {(x i , t i , y i )}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "Our aim is to develop a sequence classification model which considers the temporal information t i along with x i for classifying a social media post. In particular, we consider the rumour stance classification problem in Twitter where one classifies tweets into Support, Query, Deny, and Comment class, thus y i \u2208 Y={Support, Query, Deny, Comment}.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Background",
"sec_num": "2"
},
{
"text": "NODE were introduced as a continuous depth alternative to Residual Networks (ResNets) (He et al., 2016) . ResNets uses skip connections to avoid vanishing gradient problems when networks grow deeper. Residual block output is computed as",
"cite_spans": [
{
"start": 86,
"end": 103,
"text": "(He et al., 2016)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Neural Ordinary Differential Equations",
"sec_num": "2.1"
},
{
"text": "h t+1 = h t + f (h t , \u03b8 t ),",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Neural Ordinary Differential Equations",
"sec_num": "2.1"
},
{
"text": "where f () is a neural network (NN) parameterized by \u03b8 t and h t representing the hidden representation at depth t. This update is similar to a step in Euler numerical technique used for solving ordinary differential equations (ODE) dh(t) dt = f (h(t), t, \u03b8). The sequence of residual block operations in ResNets can be seen as a solution to this ODE. Consequently, NODEs can be interpreted as a continuous equivalent of ResNets modeling the evolution of hidden representationsh(t) over time.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Neural Ordinary Differential Equations",
"sec_num": "2.1"
},
{
"text": "For solving ODE, one can use fixed stepsize numerical techniques such as Euler, Runge-Kutta or adaptive step-size methods like Do-pri5(Dormand and Prince, 1980) . Solving an ODE requires one to specify an initial value h(0) (input x or its transformation) and can compute the value at t using an ODE solver ODESolverCompute(f \u03b8 , h(0), 0, t). An ODE is solved until some end-time T to obtain the final hidden representation h(T ) which is used to predict class labels\u0177. For classification problems, cross-entropy loss is used and parameters are learnt through adjoint sensitivity method (Zhuang et al., 2020; Chen et al., 2018) which provides efficient back-propagation and gradient computations.",
"cite_spans": [
{
"start": 127,
"end": 160,
"text": "Do-pri5(Dormand and Prince, 1980)",
"ref_id": null
},
{
"start": 587,
"end": 608,
"text": "(Zhuang et al., 2020;",
"ref_id": "BIBREF16"
},
{
"start": 609,
"end": 627,
"text": "Chen et al., 2018)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Neural Ordinary Differential Equations",
"sec_num": "2.1"
},
{
"text": "LSTMs are popular for sequence classification but only considers the sequential nature of the data and ignore the temporal features associated with the data in its standard setting. As the posts occur in irregular intervals of time, the nature of a new post will be influenced by the recent posts, influence will be inversely proportional to the time gap. In these situations, it will be beneficial to use a model where the number of transformations depend on the time gap.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bi-Directional Recurrent NODE",
"sec_num": "3"
},
{
"text": "We propose to use RNODE which considers the arrival time and accordingly the hidden representations are transformed across time. In RNODE, the transformation of a hidden representation h(t i\u22121 ) at time t i\u22121 to h(t i ) at time t i is governed by an ODE parameterized by a NN f (). Unlike standard LSTMs where h(t i ) is obtained from h(t i\u22121 ) as a single NN transformation, RNODE first obtains a hidden representation h \u2032 (t i ) as a solution to an ODE at time t i with initial value h(t i\u22121 ). The number of update steps in the numerical technique used to solve this ODE depends on the time gap t i \u2212t i\u22121 between the consecutive posts. The hidden representation h \u2032 (t i ) and input post x i at time t i are passed through neural network transformation (RNNCell()) to obtain final hidden representation",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bi-Directional Recurrent NODE",
"sec_num": "3"
},
{
"text": "h(t i ), i.e., h(t i ) = RNNCell(h \u2032 (t i ), x i ).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Bi-Directional Recurrent NODE",
"sec_num": "3"
},
{
"text": "The process is repeated for every element (x i , t i ) in the sequence. The hidden representations associated with the elements in the sequence are then passed to a neural network (NN()) to obtain the post labels. Using standard cross-entropy loss, the parameters of the models are learnt through backpropagation. Figure 1 provides the detailed architecture of the Bi-directional RNNs (Schuster and Paliwal, 1997 ) such as Bi-LSTMS (Graves et al., 2013) were proven to be successful in many sequence labeling tasks in natural language processing such as POS tagging (Huang et al., 2015) . They use the information from the past and future to predict the label while standard LSTMs consider only information from the past. We propose a Bi-RNODE model, which uses the sequence of input observations from past and from the future to predict the post label at any time t. It assumes the hidden representation dynamics are influenced not only by the past posts but also by the futures posts. Unlike Bi-LSTMs, Bi-RNODE considers the exact time of the posts and their inter-arrival times in determining the transformations in the hidden representations. Bi-RNODE consists of two RNODE blocks, one performing transformations in the forward direction (in the order of posting times) and the other in the backward direction. The hidden representations H and H b computed by forward and backward RNODE respectively are aggregated either by concatenation or by averaging appropriately to obtain a final hidden representation and is passed through a NN to obtain the post labels. Bi-RNODE is useful when a sequence of posts with their time of occurrence needs to be classified together. Figure 2 provides an overview of Bi-RNODE model for post classification. For Bi-RNODE, an extra neural network f \u03b8 \u2032 () is required to compute hidden representations h b (t \u2032 i ) in the backward direction. Training in Bi-RNODE is done in a similar manner to RNODE, with cross-entropy loss and back-propagation to estimate parameters.",
"cite_spans": [
{
"start": 385,
"end": 412,
"text": "(Schuster and Paliwal, 1997",
"ref_id": "BIBREF14"
},
{
"start": 432,
"end": 453,
"text": "(Graves et al., 2013)",
"ref_id": "BIBREF6"
},
{
"start": 566,
"end": 586,
"text": "(Huang et al., 2015)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [
{
"start": 314,
"end": 322,
"text": "Figure 1",
"ref_id": "FIGREF0"
},
{
"start": 1674,
"end": 1682,
"text": "Figure 2",
"ref_id": "FIGREF1"
}
],
"eq_spans": [],
"section": "Bi-Directional Recurrent NODE",
"sec_num": "3"
},
{
"text": "To demonstrate the effectiveness of the proposed approaches, we consider the stance classification problem in Twitter and RumourEval-2019 (Derczynski et al., 2019 data set. This Twitter data set consists of rumours associated with eight events. Each event has a collection of tweets labelled with one of the four labels -Support, Query, Deny and Comment. We picked four major events Charliehebdo, Ferguson, Ottawashooting and Sydneysiege (each with approximately 1000 tweets per event) from RumourEval-2019 to perform experiments.",
"cite_spans": [
{
"start": 122,
"end": 137,
"text": "RumourEval-2019",
"ref_id": null
},
{
"start": 138,
"end": 162,
"text": "(Derczynski et al., 2019",
"ref_id": null
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "Features : For dataset preparation, each data point x i associated with a Tweet includes text embedding, retweet count, favourites count, punctuation features, negative and positive word count, presence of hashtags, user mentions, URLs etc. obtained from the tweet. The text embedding of the tweet is obtained by concatenating the word embeddings 2 . Each tweet timestamp is converted to epoch time and Min-Max normalization is applied over the time stamps associated with each event to keep the duration of the event in the interval [0, 1].",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experiments",
"sec_num": "4"
},
{
"text": "We conducted experiments to predict the stance of social media posts propagating in seen events and unseen events.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "4.1"
},
{
"text": "-Seen Event Here we train, validate and test on tweets of the same event. Each event data is split 60:20:20 ratio in sequence of time. This setup helps in predicting the stance of unseen tweets of the same event.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "4.1"
},
{
"text": "-Unseen Event: This setup helps in evaluating performance on an unseen event and training on a larger dataset. Here, training and validation data are formed using data from 3 events and testing is done on the 4 th event. Last 20% of the training data (after ordering based on time) are set aside for validation. During training, mini-batches are formed only from the tweets belonging to the same event.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "4.1"
},
{
"text": "Baselines: We compared results of our proposed RNODE and Bi-RNODE models with RNN based baselines such LSTM (Kochkina et al., 2017) , Bi-LSTM (Augenstein et al., 2016) , GRU (Cho et al., 2014) , Bi-GRU, and Majority (labelling with most frequent class) baseline models. We also use a variant of LSTM baseline considering temporal information (Zubiaga et al., 2018b) , LSTM-timeGap where the time gap of consecutive data points is included as part of the input data.",
"cite_spans": [
{
"start": 108,
"end": 131,
"text": "(Kochkina et al., 2017)",
"ref_id": "BIBREF9"
},
{
"start": 142,
"end": 167,
"text": "(Augenstein et al., 2016)",
"ref_id": "BIBREF0"
},
{
"start": 174,
"end": 192,
"text": "(Cho et al., 2014)",
"ref_id": "BIBREF2"
},
{
"start": 342,
"end": 365,
"text": "(Zubiaga et al., 2018b)",
"ref_id": "BIBREF19"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "4.1"
},
{
"text": "Evaluation Metrics: We consider the standard evaluation metrics such as precision, recall, F1 and in addition the AUC score to account for the data imbalance. We consider a weighted average of the Table 1 : Performance of all the models on RumourEval-2019 (Derczynski et al., 2019) dataset. First and second rows of each model represents seen event and unseen event experiment results respectively. evaluation metrics to compare the performance of models. Hyperparameters: All the models are trained for 50 epochs with 0.01 learning rate, Adam optimizer, dropout(0.2) regularizer, batchsize of 50, hidden representation size of 64 and cross entropy as the loss function. Different hyperparameters like neural network layers (1, 2), numerical methods (Euler, RK4, Dopri5 for RNODE and Bi-RNODE) and aggregation strategy (concatenation or averaging for Bi-LSTM Bi-GRU and Bi-RNODE) are used for all the models and the best configuration is selected from the validation data for different experimental setups and train/test data splits.",
"cite_spans": [
{
"start": 256,
"end": 281,
"text": "(Derczynski et al., 2019)",
"ref_id": null
}
],
"ref_spans": [
{
"start": 197,
"end": 204,
"text": "Table 1",
"ref_id": null
}
],
"eq_spans": [],
"section": "Experimental setup",
"sec_num": "4.1"
},
{
"text": "The results of seen event and unseen event experiment setup can be found in Table 1 , where the first and second rows for each model provides results on seen event and unseen event respectively. We can observe from Table 1 that for both seen event and unseen event experiment setup, RNODE and Bi-RNODE models performed better than the baseline models in general for all the 3 events 3 . In particular for the seen event setup, Bi-RNODE gives the best result outperforming RNODE and other models for most of the data sets and measures. Under seen event experiment on Syndneysiege event, we plot the ROC curve for all the models in Figure 3 . We can observe that AUC for Figures 3(a) and 3(e) corresponding to RNODE and Bi-RNODE respectively are higher than LSTM, GRU, Bi-LSTM , and Bi-GRU.",
"cite_spans": [],
"ref_spans": [
{
"start": 76,
"end": 83,
"text": "Table 1",
"ref_id": null
},
{
"start": 215,
"end": 222,
"text": "Table 1",
"ref_id": null
},
{
"start": 630,
"end": 638,
"text": "Figure 3",
"ref_id": null
}
],
"eq_spans": [],
"section": "Results and Analysis",
"sec_num": "4.2"
},
{
"text": "We proposed RNODE, Bi-RNODE models for sequence classification of social media posts. These models consider temporal information of the posts and hidden representation are evolved as solution to ODE. Through experiments, we show these models perform better than LSTMs on rumour stance classification problem in Twitter",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusion",
"sec_num": "5"
},
{
"text": "Using pre-trained word2vec vectors which are trained on Google News dataset: https://code.google.com/p/word2vec, each word is represented as an embedding of size 15.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "Due to space constraint,Table 1presents results for 3 events, Syndneysiege results inFigure 3.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Stance detection with bidirectional conditional encoding",
"authors": [
{
"first": "Isabelle",
"middle": [],
"last": "Augenstein",
"suffix": ""
},
{
"first": "Tim",
"middle": [],
"last": "Rockt\u00e4schel",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "876--885",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Isabelle Augenstein, Tim Rockt\u00e4schel, Andreas Vla- chos, and Kalina Bontcheva. 2016. Stance detection with bidirectional conditional encoding. In Proceed- ings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 876-885.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "Neural ordinary differential equations",
"authors": [
{
"first": "T",
"middle": [
"Q"
],
"last": "Ricky",
"suffix": ""
},
{
"first": "Yulia",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "Jesse",
"middle": [],
"last": "Rubanova",
"suffix": ""
},
{
"first": "David",
"middle": [
"K"
],
"last": "Bettencourt",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Duvenaud",
"suffix": ""
}
],
"year": 2018,
"venue": "Advances in neural information processing systems",
"volume": "",
"issue": "",
"pages": "6571--6583",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Ricky TQ Chen, Yulia Rubanova, Jesse Bettencourt, and David K Duvenaud. 2018. Neural ordinary differen- tial equations. In Advances in neural information processing systems, pages 6571-6583.",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "On the properties of neural machine translation: Encoder-decoder approaches",
"authors": [
{
"first": "Kyunghyun",
"middle": [],
"last": "Cho",
"suffix": ""
},
{
"first": "Bart",
"middle": [],
"last": "Van Merri\u00ebnboer",
"suffix": ""
},
{
"first": "Dzmitry",
"middle": [],
"last": "Bahdanau",
"suffix": ""
},
{
"first": "Yoshua",
"middle": [],
"last": "Bengio",
"suffix": ""
}
],
"year": 2014,
"venue": "Proceedings of SSST-8, Eighth Workshop on Syntax, Semantics and Structure in Statistical Translation",
"volume": "",
"issue": "",
"pages": "103--111",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kyunghyun Cho, Bart van Merri\u00ebnboer, Dzmitry Bah- danau, and Yoshua Bengio. 2014. On the properties of neural machine translation: Encoder-decoder ap- proaches. In Proceedings of SSST-8, Eighth Work- shop on Syntax, Semantics and Structure in Statistical Translation, pages 103-111.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Topical stance detection for twitter: A twophase lstm model using attention",
"authors": [
{
"first": "Kuntal",
"middle": [],
"last": "Dey",
"suffix": ""
},
{
"first": "Ritvik",
"middle": [],
"last": "Shrivastava",
"suffix": ""
},
{
"first": "Saroj",
"middle": [],
"last": "Kaushik",
"suffix": ""
}
],
"year": 2018,
"venue": "Advances in Information Retrieval",
"volume": "",
"issue": "",
"pages": "529--536",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kuntal Dey, Ritvik Shrivastava, and Saroj Kaushik. 2018. Topical stance detection for twitter: A two- phase lstm model using attention. In Advances in Information Retrieval, pages 529-536.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "A family of embedded runge-kutta formulae",
"authors": [
{
"first": "R",
"middle": [],
"last": "John",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Dormand",
"suffix": ""
},
{
"first": "J",
"middle": [],
"last": "Peter",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Prince",
"suffix": ""
}
],
"year": 1980,
"venue": "Journal of computational and applied mathematics",
"volume": "6",
"issue": "1",
"pages": "19--26",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "John R Dormand and Peter J Prince. 1980. A family of embedded runge-kutta formulae. Journal of compu- tational and applied mathematics, 6(1):19-26.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "Speech recognition with deep recurrent neural networks",
"authors": [
{
"first": "Alex",
"middle": [],
"last": "Graves",
"suffix": ""
},
{
"first": "Mohamed",
"middle": [],
"last": "Abdel-Rahman",
"suffix": ""
},
{
"first": "Geoffrey",
"middle": [],
"last": "Hinton",
"suffix": ""
}
],
"year": 2013,
"venue": "2013 IEEE International Conference on Acoustics, Speech and Signal Processing",
"volume": "",
"issue": "",
"pages": "6645--6649",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Alex Graves, Abdel-rahman Mohamed, and Geoffrey Hinton. 2013. Speech recognition with deep recur- rent neural networks. In 2013 IEEE International Conference on Acoustics, Speech and Signal Process- ing, pages 6645-6649.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Deep residual learning for image recognition",
"authors": [
{
"first": "Kaiming",
"middle": [],
"last": "He",
"suffix": ""
},
{
"first": "Xiangyu",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Shaoqing",
"middle": [],
"last": "Ren",
"suffix": ""
},
{
"first": "Jian",
"middle": [],
"last": "Sun",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of the IEEE conference on computer vision and pattern recognition",
"volume": "",
"issue": "",
"pages": "770--778",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. 2016. Deep residual learning for image recog- nition. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 770- 778.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Bidirectional LSTM-CRF models for sequence tagging",
"authors": [
{
"first": "Zhiheng",
"middle": [],
"last": "Huang",
"suffix": ""
},
{
"first": "W",
"middle": [],
"last": "Xu",
"suffix": ""
},
{
"first": "Kai",
"middle": [],
"last": "Yu",
"suffix": ""
}
],
"year": 2015,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Zhiheng Huang, W. Xu, and Kai Yu. 2015. Bidirectional LSTM-CRF models for sequence tagging. ArXiv abs/1508.01991.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Turing at SemEval-2017 task 8: Sequential approach to rumour stance classification with branch-LSTM",
"authors": [
{
"first": "Elena",
"middle": [],
"last": "Kochkina",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Liakata",
"suffix": ""
},
{
"first": "Isabelle",
"middle": [],
"last": "Augenstein",
"suffix": ""
}
],
"year": 2017,
"venue": "Proceedings of the 11th International Workshop on Semantic Evaluation",
"volume": "",
"issue": "",
"pages": "475--480",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Elena Kochkina, Maria Liakata, and Isabelle Augen- stein. 2017. Turing at SemEval-2017 task 8: Sequen- tial approach to rumour stance classification with branch-LSTM. In Proceedings of the 11th Interna- tional Workshop on Semantic Evaluation (SemEval- 2017), pages 475-480.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Towards early identification of online rumors based on long short-term memory networks",
"authors": [
{
"first": "Y",
"middle": [],
"last": "Liu",
"suffix": ""
},
{
"first": "X",
"middle": [],
"last": "Jin",
"suffix": ""
},
{
"first": "H",
"middle": [],
"last": "Shen",
"suffix": ""
}
],
"year": 2019,
"venue": "Inf. Process. Manag",
"volume": "56",
"issue": "",
"pages": "1457--1467",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Y. Liu, X. Jin, and H. Shen. 2019. Towards early iden- tification of online rumors based on long short-term memory networks. Inf. Process. Manag., 56:1457- 1467.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Gaussian processes for rumour stance classification in social media",
"authors": [
{
"first": "Michal",
"middle": [],
"last": "Lukasik",
"suffix": ""
},
{
"first": "Kalina",
"middle": [],
"last": "Bontcheva",
"suffix": ""
},
{
"first": "Trevor",
"middle": [],
"last": "Cohn",
"suffix": ""
},
{
"first": "Arkaitz",
"middle": [],
"last": "Zubiaga",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Liakata",
"suffix": ""
},
{
"first": "Rob",
"middle": [],
"last": "Procter",
"suffix": ""
}
],
"year": 2019,
"venue": "ACM Trans. Inf. Syst",
"volume": "37",
"issue": "2",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Michal Lukasik, Kalina Bontcheva, Trevor Cohn, Arkaitz Zubiaga, Maria Liakata, and Rob Procter. 2019. Gaussian processes for rumour stance classifi- cation in social media. ACM Trans. Inf. Syst., 37(2).",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Rumor has it: Identifying misinformation in microblogs",
"authors": [
{
"first": "Emily",
"middle": [],
"last": "Vahed Qazvinian",
"suffix": ""
},
{
"first": "Dragomir",
"middle": [],
"last": "Rosengren",
"suffix": ""
},
{
"first": "Qiaozhu",
"middle": [],
"last": "Radev",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Mei",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1589--1599",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Vahed Qazvinian, Emily Rosengren, Dragomir Radev, and Qiaozhu Mei. 2011. Rumor has it: Identifying misinformation in microblogs. In Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing, pages 1589-1599.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Latent odes for irregularly-sampled time series",
"authors": [
{
"first": "Yulia",
"middle": [],
"last": "Rubanova",
"suffix": ""
},
{
"first": "T",
"middle": [
"Q"
],
"last": "Ricky",
"suffix": ""
},
{
"first": "David",
"middle": [],
"last": "Chen",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Duvenaud",
"suffix": ""
}
],
"year": 2019,
"venue": "Proceedings of the 33rd International Conference on Neural Information Processing Systems",
"volume": "",
"issue": "",
"pages": "5320--5330",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Yulia Rubanova, Ricky TQ Chen, and David Duvenaud. 2019. Latent odes for irregularly-sampled time series. In Proceedings of the 33rd International Conference on Neural Information Processing Systems, pages 5320-5330.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Bidirectional recurrent neural networks",
"authors": [
{
"first": "Mike",
"middle": [],
"last": "Schuster",
"suffix": ""
},
{
"first": "K",
"middle": [],
"last": "Kuldip",
"suffix": ""
},
{
"first": "",
"middle": [],
"last": "Paliwal",
"suffix": ""
}
],
"year": 1997,
"venue": "IEEE transactions on Signal Processing",
"volume": "45",
"issue": "11",
"pages": "2673--2681",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mike Schuster and Kuldip K Paliwal. 1997. Bidirec- tional recurrent neural networks. IEEE transactions on Signal Processing, 45(11):2673-2681.",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Early detection of rumours on twitter via stance transfer learning",
"authors": [
{
"first": "Lin",
"middle": [],
"last": "Tian",
"suffix": ""
},
{
"first": "Xiuzhen",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Yan",
"middle": [],
"last": "Wang",
"suffix": ""
},
{
"first": "Huan",
"middle": [],
"last": "Liu",
"suffix": ""
}
],
"year": 2020,
"venue": "European Conference on Information Retrieval",
"volume": "",
"issue": "",
"pages": "575--588",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Lin Tian, Xiuzhen Zhang, Yan Wang, and Huan Liu. 2020. Early detection of rumours on twitter via stance transfer learning. In European Conference on Information Retrieval, pages 575-588. Springer.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Adaptive checkpoint adjoint method for gradient estimation in neural ode",
"authors": [
{
"first": "Juntang",
"middle": [],
"last": "Zhuang",
"suffix": ""
},
{
"first": "Nicha",
"middle": [],
"last": "Dvornek",
"suffix": ""
},
{
"first": "Xiaoxiao",
"middle": [],
"last": "Li",
"suffix": ""
},
{
"first": "Sekhar",
"middle": [],
"last": "Tatikonda",
"suffix": ""
}
],
"year": 2020,
"venue": "International Conference on Machine Learning",
"volume": "",
"issue": "",
"pages": "11639--11649",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Juntang Zhuang, Nicha Dvornek, Xiaoxiao Li, Sekhar Tatikonda, Xenophon Papademetris, and James Dun- can. 2020. Adaptive checkpoint adjoint method for gradient estimation in neural ode. In Inter- national Conference on Machine Learning, pages 11639-11649. PMLR.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Detection and resolution of rumours in social media: A survey",
"authors": [
{
"first": "Arkaitz",
"middle": [],
"last": "Zubiaga",
"suffix": ""
},
{
"first": "Ahmet",
"middle": [],
"last": "Aker",
"suffix": ""
},
{
"first": "Kalina",
"middle": [],
"last": "Bontcheva",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Liakata",
"suffix": ""
},
{
"first": "Rob",
"middle": [],
"last": "Procter",
"suffix": ""
}
],
"year": 2018,
"venue": "ACM Computing Surveys (CSUR)",
"volume": "51",
"issue": "2",
"pages": "1--36",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Arkaitz Zubiaga, Ahmet Aker, Kalina Bontcheva, Maria Liakata, and Rob Procter. 2018a. Detection and res- olution of rumours in social media: A survey. ACM Computing Surveys (CSUR), 51(2):1-36.",
"links": null
},
"BIBREF18": {
"ref_id": "b18",
"title": "Stance classification in rumours as a sequential task exploiting the tree structure of social media conversations",
"authors": [
{
"first": "Arkaitz",
"middle": [],
"last": "Zubiaga",
"suffix": ""
},
{
"first": "Elena",
"middle": [],
"last": "Kochkina",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Liakata",
"suffix": ""
},
{
"first": "Rob",
"middle": [],
"last": "Procter",
"suffix": ""
},
{
"first": "Michal",
"middle": [],
"last": "Lukasik",
"suffix": ""
}
],
"year": 2016,
"venue": "Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers",
"volume": "",
"issue": "",
"pages": "2438--2448",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Arkaitz Zubiaga, Elena Kochkina, Maria Liakata, Rob Procter, and Michal Lukasik. 2016. Stance classi- fication in rumours as a sequential task exploiting the tree structure of social media conversations. In Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 2438-2448.",
"links": null
},
"BIBREF19": {
"ref_id": "b19",
"title": "Discourseaware rumour stance classification in social media using sequential classifiers",
"authors": [
{
"first": "Arkaitz",
"middle": [],
"last": "Zubiaga",
"suffix": ""
},
{
"first": "Elena",
"middle": [],
"last": "Kochkina",
"suffix": ""
},
{
"first": "Maria",
"middle": [],
"last": "Liakata",
"suffix": ""
},
{
"first": "Rob",
"middle": [],
"last": "Procter",
"suffix": ""
},
{
"first": "Michal",
"middle": [],
"last": "Lukasik",
"suffix": ""
},
{
"first": "Kalina",
"middle": [],
"last": "Bontcheva",
"suffix": ""
},
{
"first": "Trevor",
"middle": [],
"last": "Cohn",
"suffix": ""
},
{
"first": "Isabelle",
"middle": [],
"last": "Augenstein",
"suffix": ""
}
],
"year": 2018,
"venue": "Information Processing & Management",
"volume": "54",
"issue": "2",
"pages": "273--290",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Arkaitz Zubiaga, Elena Kochkina, Maria Liakata, Rob Procter, Michal Lukasik, Kalina Bontcheva, Trevor Cohn, and Isabelle Augenstein. 2018b. Discourse- aware rumour stance classification in social media using sequential classifiers. Information Processing & Management, 54(2):273-290.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"num": null,
"text": "Architecture details of RNODE",
"type_str": "figure",
"uris": null
},
"FIGREF1": {
"num": null,
"text": "Bi-RNODE Architecture RNODE model.",
"type_str": "figure",
"uris": null
},
"TABREF0": {
"content": "<table><tr><td>Model</td><td colspan=\"2\">Charliehebdo</td><td/><td colspan=\"2\">Ferguson</td><td/><td colspan=\"2\">Ottawashooting</td></tr><tr><td/><td>AUC F1</td><td colspan=\"3\">Recall Preci-AUC F1</td><td colspan=\"3\">Recall Preci-AUC F1</td><td>Recall Preci-</td></tr><tr><td/><td/><td/><td>sion</td><td/><td/><td>sion</td><td/><td>sion</td></tr><tr><td>RNODE</td><td colspan=\"2\">0.665 0.653 0.674</td><td>0.658</td><td>0.600 0.591</td><td>0.659</td><td>0.598</td><td colspan=\"2\">0.638 0.654 0.692</td><td>0.670</td></tr><tr><td/><td colspan=\"2\">0.638 0.672 0.700</td><td>0.721</td><td>0.618 0.632</td><td>0.677</td><td>0.640</td><td colspan=\"2\">0.659 0.651 0.703</td><td>0.642</td></tr><tr><td colspan=\"3\">Bi-RNODE 0.696 0.659 0.693</td><td>0.629</td><td>0.595 0.599</td><td>0.673</td><td>0.641</td><td colspan=\"2\">0.669 0.667 0.692</td><td>0.658</td></tr><tr><td/><td colspan=\"2\">0.651 0.697 0.737</td><td>0.690</td><td>0.615 0.643</td><td>0.695</td><td>0.635</td><td colspan=\"2\">0.652 0.624 0.662</td><td>0.618</td></tr><tr><td>Bi-LSTM</td><td colspan=\"2\">0.628 0.625 0.679</td><td>0.609</td><td>0.563 0.599</td><td>0.650</td><td>0.614</td><td colspan=\"2\">0.622 0.627 0.654</td><td>0.622</td></tr><tr><td/><td colspan=\"2\">0.662 0.690 0.717</td><td>0.671</td><td>0.603 0.623</td><td>0.667</td><td>0.600</td><td colspan=\"2\">0.650 0.637 0.686</td><td>0.622</td></tr><tr><td>Bi-GRU</td><td colspan=\"2\">0.654 0.643 0.660</td><td>0.641</td><td>0.588 0.571</td><td>0.631</td><td>0.625</td><td colspan=\"2\">0.640 0.651 0.686</td><td>0.644</td></tr><tr><td/><td colspan=\"2\">0.656 0.690 0.724</td><td>0.682</td><td>0.613 0.634</td><td>0.678</td><td>0.611</td><td colspan=\"2\">0.648 0.636 0.683</td><td>0.610</td></tr><tr><td>LSTM</td><td colspan=\"2\">0.625 0.600 0.637</td><td>0.637</td><td>0.567 0.602</td><td>0.650</td><td>0.611</td><td colspan=\"2\">0.605 0.609 0.635</td><td>0.603</td></tr><tr><td/><td colspan=\"2\">0.645 0.690 0.728</td><td>0.686</td><td>0.602 0.611</td><td>0.631</td><td>0.603</td><td colspan=\"2\">0.630 0.626 0.680</td><td>0.627</td></tr><tr><td>GRU</td><td colspan=\"2\">0.616 0.610 0.647</td><td>0.623</td><td>0.578 0.588</td><td>0.664</td><td>0.631</td><td colspan=\"2\">0.591 0.539 0.513</td><td>0.574</td></tr><tr><td/><td colspan=\"2\">0.682 0.695 0.713</td><td>0.686</td><td>0.614 0.640</td><td>0.687</td><td>0.623</td><td colspan=\"2\">0.638 0.632 0.683</td><td>0.618</td></tr><tr><td>LSTM-</td><td colspan=\"2\">0.638 0.631 0.679</td><td>0.605</td><td>0.565 0.581</td><td>0.627</td><td>0.590</td><td colspan=\"2\">0.625 0.640 0.679</td><td>0.650</td></tr><tr><td>timeGap</td><td colspan=\"2\">0.652 0.695 0.732</td><td>0.696</td><td>0.604 0.625</td><td>0.673</td><td>0.633</td><td colspan=\"2\">0.638 0.638 0.683</td><td>0.651</td></tr><tr><td>Majority</td><td colspan=\"2\">0.500 0.456 0.605</td><td>0.366</td><td>0.500 0.518</td><td>0.654</td><td>0.428</td><td colspan=\"2\">0.500 0.485 0.628</td><td>0.395</td></tr><tr><td/><td colspan=\"2\">0.500 0.542 0.673</td><td>0.453</td><td>0.500 0.528</td><td>0.662</td><td>0.439</td><td colspan=\"2\">0.500 0.467 0.614</td><td>0.377</td></tr></table>",
"html": null,
"text": "Figure 3: ROC curves of different models trained on sydneysiege event for seen event experimental setup. Bi-RNODE exhibits better AUC and class separability overall classes.",
"type_str": "table",
"num": null
}
}
}
} |