File size: 54,698 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 | {
"paper_id": "S13-1018",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T15:42:30.738205Z"
},
"title": "UBC UOS-TYPED: Regression for Typed-similarity",
"authors": [
{
"first": "Eneko",
"middle": [],
"last": "Agirre",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of the Basque Country Donostia",
"location": {
"postCode": "20018",
"country": "Basque Country"
}
},
"email": "e.agirre@ehu.es"
},
{
"first": "Nikolaos",
"middle": [],
"last": "Aletras",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield Sheffield",
"location": {
"postCode": "S1 4DP",
"country": "UK"
}
},
"email": "n.aletras@dcs.shef.ac.uk"
},
{
"first": "Aitor",
"middle": [],
"last": "Gonzalez-Agirre",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of the Basque Country Donostia",
"location": {
"postCode": "20018",
"settlement": "Basque Country"
}
},
"email": ""
},
{
"first": "German",
"middle": [],
"last": "Rigau",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of the Basque Country Donostia",
"location": {
"postCode": "20018",
"settlement": "Basque Country"
}
},
"email": "german.rigau@ehu.es"
},
{
"first": "Mark",
"middle": [],
"last": "Stevenson",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "University of Sheffield Sheffield",
"location": {
"postCode": "S1 4DP",
"country": "UK"
}
},
"email": "m.stevenson@dcs.shef.ac.uk"
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "We approach the typed-similarity task using a range of heuristics that rely on information from the appropriate metadata fields for each type of similarity. In addition we train a linear regressor for each type of similarity. The results indicate that the linear regression is key for good performance. Our best system was ranked third in the task.",
"pdf_parse": {
"paper_id": "S13-1018",
"_pdf_hash": "",
"abstract": [
{
"text": "We approach the typed-similarity task using a range of heuristics that rely on information from the appropriate metadata fields for each type of similarity. In addition we train a linear regressor for each type of similarity. The results indicate that the linear regression is key for good performance. Our best system was ranked third in the task.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "The typed-similarity dataset comprises pairs of Cultural Heritage items from Europeana 1 , a single access point to digitised versions of books, paintings, films, museum objects and archival records from institutions throughout Europe. Typically, the items comprise meta-data describing a cultural heritage item and, sometimes, a thumbnail of the item itself. Participating systems need to compute the similarity between items using the textual meta-data. In addition to general similarity, the dataset includes specific kinds of similarity, like similar author, similar time period, etc.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We approach the problem using a range of similarity techniques for each similarity types, these make use of information contained in the relevant meta-data fields.In addition, we train a linear regressor for each type of similarity, using the training data provided by the organisers with the previously defined similarity measures as features.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "We begin by describing our basic system in Section 2, followed by the machine learning system in 1 http://www.europeana.eu/ Section 3. The submissions are explained in Section 4. Section 5 presents our results. Finally, we draw our conclusions in Section 6.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "The items in this task are taken from Europeana. They cannot be redistributed, so we used the urls and scripts provided by the organizers to extract the corresponding metadata. We analysed the text in the metadata, performing lemmatization, PoS tagging, named entity recognition and classification (NERC) and date detection using Stanford CoreNLP (Finkel et al., 2005; Toutanova et al., 2003) . A preliminary score for each similarity type was then calculated as follows:",
"cite_spans": [
{
"start": 347,
"end": 368,
"text": "(Finkel et al., 2005;",
"ref_id": "BIBREF7"
},
{
"start": 369,
"end": 392,
"text": "Toutanova et al., 2003)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Basic system",
"sec_num": "2"
},
{
"text": "\u2022 General: cosine similarity of TF.IDF vectors of tokens, taken from all fields. \u2022 Author: cosine similarity of TF.IDF vectors of dc:Creator field. \u2022 People involved, time period and location: cosine similarity of TF.IDF vectors of location/date/people entities recognized by NERC in all fields. \u2022 Events: cosine similarity of TF.IDF vectors of event verbs and nouns. A list of verbs and nouns possibly denoting events was derived using the WordNet Morphosemantic Database 2 . \u2022 Subject and description: cosine similarity of TF.IDF vectors of respective fields. IDF values were calculated using a subset of Europeana items (the Culture Grid collection), available internally. These preliminary scores were im-proved using TF.IDF based on Wikipedia, UKB (Agirre and Soroa, 2009) and a more informed time similarity measure. We describe each of these processes in turn.",
"cite_spans": [
{
"start": 753,
"end": 777,
"text": "(Agirre and Soroa, 2009)",
"ref_id": "BIBREF0"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Basic system",
"sec_num": "2"
},
{
"text": "A common approach to computing document similarity is to represent documents as Bag-Of-Words (BOW). Each BOW is a vector consisting of the words contained in the document, where each dimension corresponds to a word, and the weight is the frequency in the corresponding document. The similarity between two documents can be computed as the cosine of the angle between their vectors. This is the approached use above.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TF.IDF",
"sec_num": "2.1"
},
{
"text": "This approach can be improved giving more weight to words which occur in only a few documents, and less weight to words occurring in many documents (Baeza-Yates and Ribeiro-Neto, 1999).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TF.IDF",
"sec_num": "2.1"
},
{
"text": "In our system, we count document frequencies of words using Wikipedia as a reference corpus since the training data consists of only 750 items associated with short textual information and might not be sufficient for reliable estimations. The TF.IDF similarity between items a and b is defined as:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TF.IDF",
"sec_num": "2.1"
},
{
"text": "sim tf.idf (a, b) = w\u2208a,b tfw,a \u00d7 tf w,b \u00d7 idf 2 w w\u2208a (tfw,a \u00d7 idfw) 2 \u00d7 w\u2208b (tf w,b \u00d7 idfw) 2",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TF.IDF",
"sec_num": "2.1"
},
{
"text": "where tf w,x is the frequency of the term w in x \u2208 {a, b} and idf w is the inverted document frequency of the word w measured in Wikipedia. We substituted the preliminary general similarity score by the obtained using the TF.IDF presented in this section.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "TF.IDF",
"sec_num": "2.1"
},
{
"text": "The semantic disambiguation UKB 3 algorithm (Agirre and Soroa, 2009) applies personalized PageRank on a graph generated from the English WordNet (Fellbaum, 1998) , or alternatively, from Wikipedia. This algorithm has proven to be very competitive in word similarity tasks (Agirre et al., 2010) . To compute similarity using UKB we represent WordNet as a graph G = (V, E) as follows: graph nodes represent WordNet concepts (synsets) and dictionary words; relations among synsets are represented by undirected edges; and dictionary words are linked to the synsets associated to them by directed edges.",
"cite_spans": [
{
"start": 44,
"end": 68,
"text": "(Agirre and Soroa, 2009)",
"ref_id": "BIBREF0"
},
{
"start": 145,
"end": 161,
"text": "(Fellbaum, 1998)",
"ref_id": "BIBREF6"
},
{
"start": 272,
"end": 293,
"text": "(Agirre et al., 2010)",
"ref_id": "BIBREF1"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "Our method is provided with a pair of vectors of words and a graph-based representation of WordNet. We first compute the personalized PageRank over WordNet separately for each of the vector of words, producing a probability distribution over WordNet synsets. We then compute the similarity between these two probability distributions by encoding them as vectors and computing the cosine between the vectors. We present each step in turn.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "Once personalized PageRank is computed, it returns a probability distribution over WordNet synsets. The similarity between two vectors of words can thus be implemented as the similarity between the probability distributions, as given by the cosine between the vectors.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "We used random walks to compute improved similarity values for author, people involved, location and event similarity:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "\u2022 Author: UKB over Wikipedia using person entities recognized by NERC in the dc:Creator field.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "\u2022 People involved and location: UKB over Wikipedia using people/location entities recognized by NERC in all fields.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "\u2022 Events: UKB over WordNet using event nouns and verbs recognized in all fields.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "Results on the training data showed that performance using this approach was quite low (with the exception of events). This was caused by the large number of cases where the Stanford parser did not find entities which were in Wikipedia. With those cases on mind, we combined the scores returned by UKB with the similarity scores presented in Section 2 as follows: if UKB similarity returns a score, we multiply both, otherwise we return the square of the other similarity score. Using the multiplication of the two scores, the results on the training data improved.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "UKB",
"sec_num": "2.2"
},
{
"text": "In order to measure the time similarity between a pair of items, we need to recognize time expressions in both items. We assume that the year of creation or the year denoting when the event took place in an artefact are good indicators for time similarity. Therefore, information about years is extracted from each item using the following pattern:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "[1|2][0 \u2212 9]{3}. Using this approach, each item is represented as a set of numbers denoting the years mentioned in the meta-data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "Time similarity between two items is computed based on the similarity between their associated years. Similarity between two years is defined as:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "sim year (y 1 , y 2 ) = max{0, 1 \u2212 |y1 \u2212 y2| * k}",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "k is a parameter to weight the difference between two years, e.g. for k = 0.1 all items that have difference of 10 years or more assigned a score of 0. We obtained best results for k = 0.1.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "Finally, time similarity between items a and b is computed as the maximum of the pairwise similarity between their associated years:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "sim time (a, b) = max \u2200i\u2208a \u2200j\u2208b {0, sim year (a i , b j )}",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "We substituted the preliminary time similarity score by the measure obtained using the method presented in this section.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Time similarity measure",
"sec_num": "2.3"
},
{
"text": "The above heuristics can be good indicators for the respective kind of similarity, and can be thus applied directly to the task. In this section, we take those indicators as features, and use linear regression (as made available by Weka (Hall et al., 2009) ) to learn models that fit the features to the training data.",
"cite_spans": [
{
"start": 237,
"end": 256,
"text": "(Hall et al., 2009)",
"ref_id": "BIBREF8"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Applying Machine Learning",
"sec_num": "3"
},
{
"text": "We generated further similarity scores for general similarity, including Latent Dirichlet Allocation (LDA) (Blei et al., 2003) , UKB and Wikipedia Link Vector Model (WLVM) (Milne, 2007) using information taken from all fields, as explained below.",
"cite_spans": [
{
"start": 107,
"end": 126,
"text": "(Blei et al., 2003)",
"ref_id": "BIBREF5"
},
{
"start": 172,
"end": 185,
"text": "(Milne, 2007)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Applying Machine Learning",
"sec_num": "3"
},
{
"text": "LDA (Blei et al., 2003) is a statistical method that learns a set of latent variables called topics from a training corpus. Given a topic model, documents can be inferred as probability distributions over topics, \u03b8. The distribution for a document i is denoted as \u03b8 i . An LDA model is trained using the training set consisting of 100 topics using the gensim package 4 . The hyperparameters (\u03b1, \u03b2) were set to 1 num of topics . Therefore, each item in the test set is represented as a topic distribution.",
"cite_spans": [
{
"start": 4,
"end": 23,
"text": "(Blei et al., 2003)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "LDA",
"sec_num": "3.1"
},
{
"text": "The similarity between a pair of items is estimated by comparing their topic distributions following the method proposed in . This is achieved by considering each distribution as a vector (consisting of the topics corresponding to an item and its probability) then computing the cosine of the angle between them, i.e.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "LDA",
"sec_num": "3.1"
},
{
"text": "sim LDA (a, b) = \u03b8 a \u2022 \u03b8 b | \u03b8 a | \u00d7 | \u03b8 b |",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "LDA",
"sec_num": "3.1"
},
{
"text": "where \u03b8 a is the vector created from the probability distribution generated by LDA for item a.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "LDA",
"sec_num": "3.1"
},
{
"text": "We run UKB (Section 2.2) to generate a probability distribution over WordNet synsets for all of the words of all items. Similarity between two words is computed by creating vectors from these distributions and comparing them using the cosine of the angle between the two vectors. If a words does not appear in WordNet its similarity value to every other word is set to 0. We refer to that similarity metric as UKB here. Similarity between two items is computed by performing pairwise comparison between their words, for each, selecting the highest similarity score:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pairwise UKB",
"sec_num": "3.2"
},
{
"text": "sim(a, b) = 1 2 w 1 \u2208a arg max w 2 \u2208b U KB(w 1 , w 2 ) |a| + w 2 \u2208b arg max w 1 \u2208a U KB(w 2 , w 1 ) |b|",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pairwise UKB",
"sec_num": "3.2"
},
{
"text": "where a and b are two items, |a| the number of tokens in a and U KB(w 1 , w 2 ) is the similarity between words w 1 and w 2 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Pairwise UKB",
"sec_num": "3.2"
},
{
"text": "An algorithm described by Milne and Witten (2008) associates Wikipedia articles which are likely to be relevant to a given text snippet using machine learning techniques. We make use of that method to represent each item as a set of likely relevant Wikipedia articles. Then, similarity between Wikipedia articles is measured using the Wikipedia Link Vector Model (WLVM) (Milne, 2007) . WLVM uses both the link structure and the article titles of Wikipedia to measure similarity between two Wikipedia articles. Each link is weighted by the probability of it occurring. Thus, the value of the weight w for a link x \u2192 y between articles x and y is:",
"cite_spans": [
{
"start": 26,
"end": 49,
"text": "Milne and Witten (2008)",
"ref_id": "BIBREF9"
},
{
"start": 370,
"end": 383,
"text": "(Milne, 2007)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "w(x \u2192 y) = |x \u2192 y| \u00d7 log t z=1 t z \u2192 y",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "where t is the total number of articles in Wikipedia. The similarity of articles is compared by forming vectors of the articles which are linked from them and computing the cosine of their angle. For example the vectors of two articles x and y are:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "x = (w(x \u2192 l 1 ), w(x \u2192 l 2 ), ..., w(x \u2192 l n )) y = (w(y \u2192 l 1 ), w(y \u2192 l 2 ), ..., w(y \u2192 l n ))",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "where x and y are two Wikipedia articles and x \u2192 l i is a link from article x to article l i .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "Since the items have been mapped to Wikipedia articles, similarity between two items is computed by performing pairwise comparison between articles using WLVM, for each, selecting the highest similarity score:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "sim(a, b) = 1 2 w 1 \u2208a arg max w 2 \u2208b W LV M (w 1 , w 2 ) |a| + w 2 \u2208b arg max w 1 \u2208a W LV M (w 2 , w 1 ) |b|",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "where a and b are two items, |a| the number of Wikipedia articles in a and W LV M (w 1 , w 2 ) is the similarity between concepts w 1 and w 2 .",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "WLVM",
"sec_num": "3.3"
},
{
"text": "We selected three systems for submission. The first run uses the similarity scores of the basic system (Section 2) for each similarity types as follows:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "\u2022 General: cosine similarity of TF.IDF vectors, IDF based on Wikipedia (as shown in Section 2.1). \u2022 Author: product of the scores obtained obtained using TF.IDF vectors and UKB (as shown in Section 2.2) using only the data extracted from dc:Creator field.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "\u2022 People involved and location: product of cosine similarity of TF.IDF vectors and UKB (as shown in Section 2.2) using the data extracted from all fields. \u2022 Time period: time similarity measure (as shown in Section 2.3). \u2022 Events: product of cosine similarity of TF.IDF vectors and UKB (as shown in Section 2.2) of event nouns and verbs recognized in all fields. \u2022 Subject and description: cosine similarity of TF.IDF vectors of respective fields (as shown in Section 2).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "For the second run we trained a ML model for each of the similarity types, using the following features:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "\u2022 Cosine similarity of TF.IDF vectors as shown in Section 2 for the eight similarity types. \u2022 Four new values for general similarity: TF.IDF (Section 2.1), LDA (Section 3.1), UKB and WLVM (Section 3.3). \u2022 Time similarity as shown in Section 2.3.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "\u2022 Events similarity computed using UKB initialized with the event nouns and verbs in all fields.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "We decided not to use the product of TF.IDF and UKB presented in Section 2.2 in this system because our intention was to measure the power of the linear regression ML algorithm to learn on the given raw data.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "The third run is similar, but includes all available features (21). In addition to the above, we included:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "\u2022 Author, people involved and location similarity computed using UKB initialized with people/location recognized by NERC in dc:Creator field for author, and in all fields for people involved and location. \u2022 Author, people involved, location and event similarity scores computed by the product of TF.IDF vectors and UKB values as shown in Section 2.2.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Submissions",
"sec_num": "4"
},
{
"text": "Evaluation was carried out using the official scorer provided by the organizers, which computes the Pearson Correlation score for each of the eight similarity types plus an additional mean correlation. ",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Results",
"sec_num": "5"
},
{
"text": "The three runs mentioned above were developed using the training data made available by the organizers. In order to avoid overfitting we did not change the default parameters of the linear regressor, and 10-fold cross-validation was used for evaluating the models on the training data. The results of our systems on the training data are shown on Table 1 . The table shows that the heuristics (RUN1) obtain low results, and that linear regression improves results considerably in all types. Using the full set of features, RUN3 improves slightly over RUN2, but the improvement is consistent across all types.",
"cite_spans": [],
"ref_spans": [
{
"start": 347,
"end": 354,
"text": "Table 1",
"ref_id": "TABREF0"
}
],
"eq_spans": [],
"section": "Development",
"sec_num": "5.1"
},
{
"text": "The test dataset was composed of 750 pairs of items. Table 2 illustrates the results of our systems in the test dataset. The results of the runs are very similar to those obtained on the training data, but the difference between RUN2 and RUN3 is even smaller. Our systems were ranked #3 (RUN 3), #4 (RUN 2) and #6 (RUN 1) among 14 systems submitted by 6 teams. Our systems achieved good correlation scores for almost all similarity types, with the exception of author similarity, which is the worst ranked in comparison with the rest of the systems.",
"cite_spans": [],
"ref_spans": [
{
"start": 53,
"end": 60,
"text": "Table 2",
"ref_id": "TABREF1"
}
],
"eq_spans": [],
"section": "Test",
"sec_num": "5.2"
},
{
"text": "In this paper, we presented the systems submitted to the *SEM 2013 shared task on Semantic Textual Similarity. We combined some simple heuristics for each type of similarity, based on the appropriate metadata fields. The use of lineal regression improved the results considerably across all types.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future Work",
"sec_num": "6"
},
{
"text": "Our system fared well in the competition. We sub-mitted three systems and the highest-ranked of these achieved the third best results overall.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Conclusions and Future Work",
"sec_num": "6"
},
{
"text": "urlhttp://wordnetcode.princeton.edu/standofffiles/morphosemantic-links.xls",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://ixa2.si.ehu.es/ukb/",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "http://pypi.python.org/pypi/gensim",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "This work is partially funded by the PATHS project (http://paths-project.eu) funded by the European Community's Seventh Framework Programme (FP7/2007-2013) under grant agreement no. 270082. Aitor Gonzalez-Agirre is supported by a PhD grant from the Spanish Ministry of Education, Culture and Sport (grant FPU12/06243).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgements",
"sec_num": null
}
],
"bib_entries": {
"BIBREF0": {
"ref_id": "b0",
"title": "Personalizing pagerank for word sense disambiguation",
"authors": [
{
"first": "Eneko",
"middle": [],
"last": "Agirre",
"suffix": ""
},
{
"first": "Aitor",
"middle": [],
"last": "Soroa",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the 12th conference of the European chapter of the Association for Computational Linguistics",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eneko Agirre and Aitor Soroa. 2009. Personalizing pagerank for word sense disambiguation. In Proceed- ings of the 12th conference of the European chapter of the Association for Computational Linguistics (EACL- 2009), Athens, Greece.",
"links": null
},
"BIBREF1": {
"ref_id": "b1",
"title": "European Language Resources Association (ELRA)",
"authors": [
{
"first": "Eneko",
"middle": [],
"last": "Agirre",
"suffix": ""
},
{
"first": "Montse",
"middle": [],
"last": "Cuadros",
"suffix": ""
},
{
"first": "German",
"middle": [],
"last": "Rigau",
"suffix": ""
},
{
"first": "Aitor",
"middle": [],
"last": "Soroa",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the Seventh conference on International Language Resources and Evaluation (LREC10)",
"volume": "",
"issue": "",
"pages": "373--377",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Eneko Agirre, Montse Cuadros, German Rigau, and Aitor Soroa. 2010. Exploring knowledge bases for sim- ilarity. In Proceedings of the Seventh conference on International Language Resources and Evaluation (LREC10). European Language Resources Associa- tion (ELRA). ISBN: 2-9517408-6-7. Pages 373-377.\".",
"links": null
},
"BIBREF2": {
"ref_id": "b2",
"title": "Computing similarity between cultural heritage items using multimodal features",
"authors": [
{
"first": "Nikolaos",
"middle": [],
"last": "Aletras",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Stevenson",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 6th Workshop on Language Technology for Cultural Heritage",
"volume": "",
"issue": "",
"pages": "85--93",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nikolaos Aletras and Mark Stevenson. 2012. Computing similarity between cultural heritage items using multi- modal features. In Proceedings of the 6th Workshop on Language Technology for Cultural Heritage, So- cial Sciences, and Humanities, pages 85-93, Avignon, France.",
"links": null
},
"BIBREF3": {
"ref_id": "b3",
"title": "Computing similarity between items in a digital library of cultural heritage",
"authors": [
{
"first": "Nikolaos",
"middle": [],
"last": "Aletras",
"suffix": ""
},
{
"first": "Mark",
"middle": [],
"last": "Stevenson",
"suffix": ""
},
{
"first": "Paul",
"middle": [],
"last": "Clough",
"suffix": ""
}
],
"year": 2012,
"venue": "J. Comput. Cult. Herit",
"volume": "5",
"issue": "4",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Nikolaos Aletras, Mark Stevenson, and Paul Clough. 2012. Computing similarity between items in a digi- tal library of cultural heritage. J. Comput. Cult. Herit., 5(4):16:1-16:19, December.",
"links": null
},
"BIBREF4": {
"ref_id": "b4",
"title": "Modern Information Retrieval",
"authors": [
{
"first": "R",
"middle": [],
"last": "Baeza-Yates",
"suffix": ""
},
{
"first": "B",
"middle": [],
"last": "Ribeiro-Neto",
"suffix": ""
}
],
"year": 1999,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "R. Baeza-Yates and B. Ribeiro-Neto. 1999. Modern In- formation Retrieval. Addison Wesley Longman Lim- ited, Essex.",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Latent Dirichlet Allocation",
"authors": [
{
"first": "David",
"middle": [
"M"
],
"last": "Blei",
"suffix": ""
},
{
"first": "Andrew",
"middle": [
"Y"
],
"last": "Ng",
"suffix": ""
},
{
"first": "Michael",
"middle": [
"I"
],
"last": "Jordan",
"suffix": ""
}
],
"year": 2003,
"venue": "Journal of Machine Learning Research",
"volume": "3",
"issue": "",
"pages": "993--1022",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "David M. Blei, Andrew Y. Ng, and Michael I. Jordan. 2003. Latent Dirichlet Allocation. Journal of Ma- chine Learning Research, 3:993-1022, March.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "WordNet: An Electronic Lexical Database",
"authors": [
{
"first": "Christiane",
"middle": [],
"last": "Fellbaum",
"suffix": ""
}
],
"year": 1998,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Christiane Fellbaum. 1998. WordNet: An Electronic Lexical Database. MIT Press.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Incorporating non-local information into information extraction systems by gibbs sampling",
"authors": [
{
"first": "Jenny",
"middle": [
"Rose"
],
"last": "Finkel",
"suffix": ""
},
{
"first": "Trond",
"middle": [],
"last": "Grenager",
"suffix": ""
},
{
"first": "Christopher",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2005,
"venue": "Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics, ACL '05",
"volume": "",
"issue": "",
"pages": "363--370",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Jenny Rose Finkel, Trond Grenager, and Christopher Manning. 2005. Incorporating non-local informa- tion into information extraction systems by gibbs sam- pling. In Proceedings of the 43rd Annual Meeting on Association for Computational Linguistics, ACL '05, pages 363-370, Stroudsburg, PA, USA. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "The weka data mining software: an update",
"authors": [
{
"first": "Mark",
"middle": [],
"last": "Hall",
"suffix": ""
},
{
"first": "Eibe",
"middle": [],
"last": "Frank",
"suffix": ""
},
{
"first": "Geoffrey",
"middle": [],
"last": "Holmes",
"suffix": ""
},
{
"first": "Bernhard",
"middle": [],
"last": "Pfahringer",
"suffix": ""
},
{
"first": "Peter",
"middle": [],
"last": "Reutemann",
"suffix": ""
},
{
"first": "Ian",
"middle": [
"H"
],
"last": "Witten",
"suffix": ""
}
],
"year": 2009,
"venue": "SIGKDD Explor. Newsl",
"volume": "11",
"issue": "1",
"pages": "10--18",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mark Hall, Eibe Frank, Geoffrey Holmes, Bernhard Pfahringer, Peter Reutemann, and Ian H. Witten. 2009. The weka data mining software: an update. SIGKDD Explor. Newsl., 11(1):10-18, November.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Learning to Link with Wikipedia",
"authors": [
{
"first": "D",
"middle": [],
"last": "Milne",
"suffix": ""
},
{
"first": "I",
"middle": [],
"last": "Witten",
"suffix": ""
}
],
"year": 2008,
"venue": "Proceedings of the ACM Conference on Information and Knowledge Management (CIKM'2008)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "D. Milne and I. Witten. 2008. Learning to Link with Wikipedia. In Proceedings of the ACM Con- ference on Information and Knowledge Management (CIKM'2008), Napa Valley, California.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Computing semantic relatedness using Wikipedia's link structure",
"authors": [
{
"first": "",
"middle": [],
"last": "Milne",
"suffix": ""
}
],
"year": 2007,
"venue": "Proceedings of the New Zealand Computer Science Research Student Conference",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Milne. 2007. Computing semantic relatedness using Wikipedia's link structure. In Proceedings of the New Zealand Computer Science Research Student Confer- ence.",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Feature-rich part-of-speech tagging with a cyclic dependency network",
"authors": [
{
"first": "Kristina",
"middle": [],
"last": "Toutanova",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Klein",
"suffix": ""
},
{
"first": "Christopher",
"middle": [
"D"
],
"last": "Manning",
"suffix": ""
},
{
"first": "Yoram",
"middle": [],
"last": "Singer",
"suffix": ""
}
],
"year": 2003,
"venue": "Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology",
"volume": "1",
"issue": "",
"pages": "173--180",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kristina Toutanova, Dan Klein, Christopher D. Manning, and Yoram Singer. 2003. Feature-rich part-of-speech tagging with a cyclic dependency network. In Pro- ceedings of the 2003 Conference of the North Ameri- can Chapter of the Association for Computational Lin- guistics on Human Language Technology -Volume 1, NAACL '03, pages 173-180, Stroudsburg, PA, USA. Association for Computational Linguistics.",
"links": null
}
},
"ref_entries": {
"TABREF0": {
"text": "Results of our systems on the training data, using cross-validation when necessary.",
"content": "<table><tr><td>Team and run</td><td colspan=\"3\">General Author People involved Time Location Event Subject Description Mean</td></tr><tr><td colspan=\"2\">UBC UOS-RUN1 0.7269 0.4474</td><td>0.4648</td><td>0.5884 0.4801 0.2522 0.4976 0.5389 0.5033</td></tr><tr><td colspan=\"2\">UBC UOS-RUN2 0.7777 0.6680</td><td>0.6767</td><td>0.7609 0.7329 0.6412 0.7516 0.8024 0.7264</td></tr><tr><td colspan=\"2\">UBC UOS-RUN3 0.7866 0.6941</td><td>0.6965</td><td>0.7654 0.7492 0.6551 0.7586 0.8067 0.7390</td></tr><tr><td>Team and run</td><td colspan=\"3\">General Author People involved Time Location Event Subject Description Mean Rank</td></tr><tr><td colspan=\"2\">UBC UOS-RUN1 0.7256 0.4568</td><td>0.4467</td><td>0.5762 0.4858 0.3090 0.5015 0.5810 0.5103 6</td></tr><tr><td colspan=\"2\">UBC UOS-RUN2 0.7457 0.6618</td><td>0.6518</td><td>0.7466 0.7244 0.6533 0.7404 0.7751 0.7124 4</td></tr><tr><td colspan=\"2\">UBC UOS-RUN3 0.7461 0.6656</td><td>0.6544</td><td>0.7411 0.7257 0.6545 0.7417 0.7763 0.7132 3</td></tr></table>",
"type_str": "table",
"html": null,
"num": null
},
"TABREF1": {
"text": "Results of our submitted systems.",
"content": "<table/>",
"type_str": "table",
"html": null,
"num": null
}
}
}
} |