File size: 55,919 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 | {
"paper_id": "D13-1003",
"header": {
"generated_with": "S2ORC 1.0.0",
"date_generated": "2023-01-19T16:42:28.026561Z"
},
"title": "Combining Generative and Discriminative Model Scores for Distant Supervision",
"authors": [
{
"first": "Benjamin",
"middle": [],
"last": "Roth",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Saarland University Spoken Language Systems Saarbr\u00fccken",
"location": {
"country": "Germany"
}
},
"email": ""
},
{
"first": "Dietrich",
"middle": [],
"last": "Klakow",
"suffix": "",
"affiliation": {
"laboratory": "",
"institution": "Saarland University Spoken Language Systems Saarbr\u00fccken",
"location": {
"country": "Germany"
}
},
"email": ""
}
],
"year": "",
"venue": null,
"identifiers": {},
"abstract": "Distant supervision is a scheme to generate noisy training data for relation extraction by aligning entities of a knowledge base with text. In this work we combine the output of a discriminative at-least-one learner with that of a generative hierarchical topic model to reduce the noise in distant supervision data. The combination significantly increases the ranking quality of extracted facts and achieves state-of-the-art extraction performance in an end-to-end setting. A simple linear interpolation of the model scores performs better than a parameter-free scheme based on nondominated sorting.",
"pdf_parse": {
"paper_id": "D13-1003",
"_pdf_hash": "",
"abstract": [
{
"text": "Distant supervision is a scheme to generate noisy training data for relation extraction by aligning entities of a knowledge base with text. In this work we combine the output of a discriminative at-least-one learner with that of a generative hierarchical topic model to reduce the noise in distant supervision data. The combination significantly increases the ranking quality of extracted facts and achieves state-of-the-art extraction performance in an end-to-end setting. A simple linear interpolation of the model scores performs better than a parameter-free scheme based on nondominated sorting.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Abstract",
"sec_num": null
}
],
"body_text": [
{
"text": "Relation extraction is the task of finding relational facts in unstructured text and putting them into a structured (tabularized) knowledge base. Training machine learning algorithms for relation extraction requires training data. If the set of relations is prespecified, the training data needs to be labeled with those relations.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Manual annotation of training data is laborious and costly, however, the knowledge base may already partially be filled with instances from the relations. This is utilized by a scheme known as distant supervision (DS) (Mintz et al., 2009) : text is automatically labeled by aligning (matching) pairs of entities that are contained in a knowledge base with their textual occurrences. Whenever such a match is encountered, the surrounding context (sentence) is assumed to express the relation.",
"cite_spans": [
{
"start": 218,
"end": 238,
"text": "(Mintz et al., 2009)",
"ref_id": "BIBREF11"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "This assumption, however, can fail. Consider the example given in (Takamatsu et al., 2012) : If the tuple",
"cite_spans": [
{
"start": 66,
"end": 90,
"text": "(Takamatsu et al., 2012)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "place_of_birth(Michael Jackson, Gary)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "is contained in the knowledge base, one matching context could be: Michael Jackson was born in Gary ...",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Introduction",
"sec_num": "1"
},
{
"text": "Michael Jackson moved from Gary ... Clearly, only the first context indeed expresses the relation and should be labeled accordingly.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "And another possible context:",
"sec_num": null
},
{
"text": "Three basic approaches have been proposed to deal with noisy distant supervision instances: The discriminative at-least-one approach , that requires that at least one of the matches for a relation-entity tuple indeed expresses the relation; The generative approach (Alfonseca et al., 2012) that separates relation-specific distributions from noise distributions by using hierarchical topic models; And the pattern correlation approach (Takamatsu et al., 2012) that assumes that contexts which match argument pairs have a high overlap in argument pairs with other patterns expressing the relation.",
"cite_spans": [
{
"start": 265,
"end": 289,
"text": "(Alfonseca et al., 2012)",
"ref_id": "BIBREF5"
},
{
"start": 435,
"end": 459,
"text": "(Takamatsu et al., 2012)",
"ref_id": "BIBREF16"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "And another possible context:",
"sec_num": null
},
{
"text": "In this work we combine 1) a discriminative atleast-one learner, that requires high scores for both a dedicated noise label and the matched relation, and 2) a generative topic model that uses a feature-based representation to separate relation-specific patterns from background or pair-specific noise. We score surface patterns and show that combining the two approaches results in a better ranking quality of relational facts. In an end-to-end evaluation we set a threshold on the pattern scores and apply the pat- terns in a TAC KBP-style evaluation. Although the surface patterns are very simple (only strings of tokens), they achieve state-of-the-art extraction results.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "And another possible context:",
"sec_num": null
},
{
"text": "The original form of distant supervision (Mintz et al., 2009) assumes all sentences containing an entity pair to be potential patterns for the relation holding between the entities. A variety of models relax this assumption and only presume that at least one of the entity pair occurrences is a textual manifestation of the relation. The first proposed model with an atleast-one learner is that of and . It consists of a factor graph that includes binary variables for contexts, and groups contexts together for each entity pair. MultiR (Hoffmann et al., 2011) can be viewed as a multi-label extension of . A further extension is MIMLRE (Surdeanu et al., 2012) , a jointly trained two-stage classification model.",
"cite_spans": [
{
"start": 41,
"end": 61,
"text": "(Mintz et al., 2009)",
"ref_id": "BIBREF11"
},
{
"start": 637,
"end": 660,
"text": "(Surdeanu et al., 2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "At-Least-One Models",
"sec_num": "2.1"
},
{
"text": "The hierarchical topic model (HierTopics) by Alfonseca et al. (2012) models the distant supervision data by a generative model. For each corpus match of an entity pair in the knowledge base, the corresponding surface pattern is assumed to be typical for either the entity pair, the relation, or neither. This principle is then used to infer distributions over patterns of one of the following types:",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hierarchical Topic Model",
"sec_num": "2.2"
},
{
"text": "1. For every entity pair, a pair-specific distribution.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hierarchical Topic Model",
"sec_num": "2.2"
},
{
"text": "2. For every relation, a relation-specific distribution.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Hierarchical Topic Model",
"sec_num": "2.2"
},
{
"text": "The generative process assumes that for each argument pair in the knowledge base, all patterns are generated by first choosing a hidden variable z which can take on three values, B for background, R for relation and P for pair. Corresponding vocabulary distributions (\u03c6 bg , \u03c6 rel , \u03c6 pair ) for generating the context patterns are chosen according to the value of z. The Dirichlet-smoothed vocabulary distributions are shared on the respective levels. Figure 1 shows the plate diagram of the HierTopics model.",
"cite_spans": [],
"ref_spans": [
{
"start": 453,
"end": 461,
"text": "Figure 1",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "A general background distribution.",
"sec_num": "3."
},
{
"text": "We use a feature-based extension (Roth and Klakow, 2013) of Alfonseca et al. (2012) to include bigrams for a more fine-grained representation of the patterns. For including features in the model, the model is extended with a second layer of hidden variables. A variable x represents a choice of B, R or P for every pattern, i.e. there is one variable x for every pattern. Each feature is generated conditioned on a second variable z \u2208 {B, R, P }, i.e. there are as many variables z for a pattern as there are features for it. First, the hidden variable x is generated, then all z variables are generated for the corresponding features (see Figure 1) . The values B, R or P of z depend on the corresponding x by a transition distribution:",
"cite_spans": [
{
"start": 33,
"end": 56,
"text": "(Roth and Klakow, 2013)",
"ref_id": "BIBREF13"
},
{
"start": 60,
"end": 83,
"text": "Alfonseca et al. (2012)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 640,
"end": 649,
"text": "Figure 1)",
"ref_id": "FIGREF0"
}
],
"eq_spans": [],
"section": "Generative Model",
"sec_num": "3.1"
},
{
"text": "P (Z i = z|X j(i) = x) = p same , if z = x 1\u2212psame 2 , otherwise",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generative Model",
"sec_num": "3.1"
},
{
"text": "where features at indices i are mapped to the corresponding pattern indices by a function j(i); p same is set to .99 to enforce the correspondence between pattern and feature topics. 1",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Generative Model",
"sec_num": "3.1"
},
{
"text": "As a second feature-based model, we employ a perceptron model that enforces constraints on the labels for patterns (Roth and Klakow, 2013) . The model consists of log-linear factors for the set of relations Algorithm 1 At-Least-One Perceptron Training",
"cite_spans": [
{
"start": 115,
"end": 138,
"text": "(Roth and Klakow, 2013)",
"ref_id": "BIBREF13"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Discriminative Model",
"sec_num": "3.2"
},
{
"text": "\u03b8 \u2190 0 for r \u2208 R do for pair \u2208 kb pairs(r) do for s \u2208 sentences(pair) do for r \u2208 R \\ r do if P (r|s, \u03b8) \u2264 P (r |s, \u03b8) then \u03b8 \u2190 \u03b8 + \u03c6(s, r) \u2212 \u03c6(s, r ) if P (N IL|s, \u03b8) \u2264 P (r |s, \u03b8) then \u03b8 \u2190 \u03b8 + \u03c6(s, N IL) \u2212 \u03c6(s, r ) if \u2200 s\u2208sentences(pair) : P (r|s, \u03b8) \u2264 P (N IL|s, \u03b8) then s * = arg maxs P (r|s,\u03b8) P (N IL|s,\u03b8) \u03b8 \u2190 \u03b8 + \u03c6(s * , r) \u2212 \u03c6(s * , N IL)",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discriminative Model",
"sec_num": "3.2"
},
{
"text": "R as well as a factor for the NIL label (no relation).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discriminative Model",
"sec_num": "3.2"
},
{
"text": "Probabilities for a relation r given a sentence pattern s are calculated by normalizing over log-linear factors defined as f",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discriminative Model",
"sec_num": "3.2"
},
{
"text": "r (s) = exp ( i \u03c6 i (s, r)\u03b8 i )",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discriminative Model",
"sec_num": "3.2"
},
{
"text": ", with \u03c6(s, r) the feature vector for sentence s and label assignment r, and \u03b8 r the feature weight vector. The learner is directed by the following semantics: First, for a sentence s that has a distant supervision match for relation r, relation r should have a higher probability than any other relation r \u2208 R \\ r. As extractions are expected to be noisy, high probabilities for N IL are enforced by a second constraint: N IL must have a higher probability than any relation r \u2208 R \\ r. Third, at least one DS sentence for an argument pair is expected to express the corresponding relation r. For sentences s for an entity pair belonging to relation r, this can be written as the following constraints: \u2200 s,r : P (r|s) > P (r |s) \u2227 P (N IL|s) > P (r |s) \u2203 s : P (r|s) > P (N IL|s) The violation of any of the above constraints triggers a perceptron update. The basic algorithm is sketched in Algorithm 1. 2",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Discriminative Model",
"sec_num": "3.2"
},
{
"text": "The per-pattern probabilities P (r|pat) are calculated as in Alfonseca et al. (2012) and aggregated over all pattern occurrences: For the topic model, the number of times the relation-specific topic has been sampled for a pattern is divided by n(pat), the number of times the same pattern has been observed. Analogously for the perceptron, the number of times a pattern co-occurs with entity pairs for r is multiplied by the perceptron score and divided by n(pat). The topic model and perceptron approaches are based on plausible yet fundamentally different principles of modeling noise without direct supervision. It is therefore an interesting question how complementary the models are and how much can be gained from a combination. As the two models do not use direct supervision, we also avoid tuning parameters for their combination.",
"cite_spans": [
{
"start": 61,
"end": 84,
"text": "Alfonseca et al. (2012)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Model Combination",
"sec_num": "3.3"
},
{
"text": "We use two schemes to obtain a combined ranking from the two model scores: The first is a ranking based on non-dominated sorting by successively computing the Pareto-frontier of the 2-dimensional score vectors (Borzsony et al., 2001; Godfrey et al., 2007) . The underlying principle is that all data points (patterns in our case) that are not dominated by another point 3 build the frontier and are ranked highest (see Figure 2) , with ties broken by linear combination. Sorting by computing the Paretofrontier has been applied to training machine translation systems (Duh et al., 2012) to combine the translation quality metrics BLEU, RIBES and NTER, each of which is based on different principles. In the context of machine translation it has been found to outperform a linear interpolation of the metrics and to be more stable to non-smooth metrics and noncomparable scalings. We compare non-dominated sorting with a simple linear interpolation with uniform weights.",
"cite_spans": [
{
"start": 210,
"end": 233,
"text": "(Borzsony et al., 2001;",
"ref_id": "BIBREF6"
},
{
"start": 234,
"end": 255,
"text": "Godfrey et al., 2007)",
"ref_id": "BIBREF8"
},
{
"start": 568,
"end": 586,
"text": "(Duh et al., 2012)",
"ref_id": "BIBREF7"
}
],
"ref_spans": [
{
"start": 419,
"end": 428,
"text": "Figure 2)",
"ref_id": "FIGREF2"
}
],
"eq_spans": [],
"section": "Model Combination",
"sec_num": "3.3"
},
{
"text": "Evaluation is done on the ranking quality according to TAC KBP gold annotations (Ji et al., 2010) of extracted facts from all TAC KBP queries from 2009-2011 and the TAC KBP 2009-2011 corpora. First, candidate sentences are retrieved in which the query entity and a second entity with the appropriate type are contained. Candidate sentences are then used to provide answer candidates if one of the extracted patterns matches. The answer candidates are ranked according to the score of the matching pattern.",
"cite_spans": [
{
"start": 80,
"end": 97,
"text": "(Ji et al., 2010)",
"ref_id": "BIBREF10"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation 4.1 Ranking-Based Evaluation",
"sec_num": "4"
},
{
"text": "The basis for pattern extraction is the noisy DS training data of a top-3 ranked system in TAC KBP 2012 (Roth et al., 2012) . The retrieval component of this system is used to obtain sentence and answer candidates (ranked according to their respective pattern scores). Evaluation results are reported as averages over per-relation results of the standard ranking metrics mean average precision (map), geometric map (gmap), precision at 5 and at 10 (p@5, p@10).",
"cite_spans": [
{
"start": 104,
"end": 123,
"text": "(Roth et al., 2012)",
"ref_id": "BIBREF14"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "Evaluation 4.1 Ranking-Based Evaluation",
"sec_num": "4"
},
{
"text": "The maximum-likelihood estimator (MLE) baseline scores patterns by the relative frequency they occur with a certain relation. The hierarchical topic (hier orig) as described in Alfonseca et al. (2012) increases the scores under most metrics, however the increase is only significant for p@5 and p@10. The feature-based extension of the topic model (hier feat) has significantly better ranking quality. Slightly better scores are obtained by the at-leastone perceptron learner. It is interesting to see that the model combinations both by non-dominated sorting perc+hier (pareto) as well as uniform interpolation perc+hier (itpl) give a further increase in ranking quality. The simpler interpolation scheme generally works best. Figure 3 shows the Precision/Recall curves of the basic models and the linear interpolation. On the P/R curve, the linear interpolation is equal or better than the single methods on all recall levels.",
"cite_spans": [
{
"start": 177,
"end": 200,
"text": "Alfonseca et al. (2012)",
"ref_id": "BIBREF5"
}
],
"ref_spans": [
{
"start": 728,
"end": 736,
"text": "Figure 3",
"ref_id": "FIGREF3"
}
],
"eq_spans": [],
"section": "Evaluation 4.1 Ranking-Based Evaluation",
"sec_num": "4"
},
{
"text": "We evaluate the extraction quality of the induced perc+hier (itpl) patterns in an end-to-end setting. We use the evaluation setting of (Surdeanu et al., 2012) and the results obtained with their pipeline for MIMLRE and their re-implementation of MultiR as a point of reference.",
"cite_spans": [
{
"start": 135,
"end": 158,
"text": "(Surdeanu et al., 2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "End-To-End Evaluation",
"sec_num": "4.2"
},
{
"text": "In Surdeanu et al. (2012) evaluation is done using a subset of queries from the TAC KBP 2010 and 2011 evaluation. The source corpus is the TAC KBP source corpus and a 2010 Wikipedia dump. Only those answers are considered in scoring that are contained in a list of possible answers from their candidates (reducing the number of gold answers from 1601 to 576 and thereby considerably increasing the value of reported recall).",
"cite_spans": [
{
"start": 3,
"end": 25,
"text": "Surdeanu et al. (2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [],
"eq_spans": [],
"section": "End-To-End Evaluation",
"sec_num": "4.2"
},
{
"text": "For evaluating our patterns, we take the same queries for testing as Surdeanu et al. (2012) . As the document collection, we use the TAC KBP source collection and a Wikipedia dump from 07/2009 that was available to us. From this document collection, we use our retrieval pipeline of Roth et al. (2012) and take those sentences that contain query entities and slot filler candidates according to NEtags. We filter out all candidates that are not contained in the list of candidates considered in (Surdeanu et al., 2012) , and use the same reduced set of 576 gold answers as the key. We tune a single threshold parameter t = .3 on held-out development data and take all patterns with higher scores. Table 2 shows that results obtained with the induced patterns compare well with state-of-the-art relation extraction systems. Table 2 : TAC Scores on (Surdeanu et al., 2012) queries. Figure 4 shows top-ranked patterns for per:title and org:top members employees, the two relations with most answers in the gold annotations. For maximum likelihood estimation the score is 1.0 if the patterns occurs only with the relation in question -this includes all cases where the pattern is only found once in the corpus. While this could be circumvented by frequency thresholding, we leave the long tail of the data as it is and let the algorithm deal with both frequent and infrequent patterns.",
"cite_spans": [
{
"start": 69,
"end": 91,
"text": "Surdeanu et al. (2012)",
"ref_id": "BIBREF15"
},
{
"start": 283,
"end": 301,
"text": "Roth et al. (2012)",
"ref_id": "BIBREF14"
},
{
"start": 495,
"end": 518,
"text": "(Surdeanu et al., 2012)",
"ref_id": "BIBREF15"
},
{
"start": 847,
"end": 870,
"text": "(Surdeanu et al., 2012)",
"ref_id": "BIBREF15"
}
],
"ref_spans": [
{
"start": 697,
"end": 704,
"text": "Table 2",
"ref_id": null
},
{
"start": 823,
"end": 830,
"text": "Table 2",
"ref_id": null
},
{
"start": 880,
"end": 888,
"text": "Figure 4",
"ref_id": "FIGREF4"
}
],
"eq_spans": [],
"section": "End-To-End Evaluation",
"sec_num": "4.2"
},
{
"text": "One can see that while the maximum likelihood patterns contain some reasonable relational contexts, they are less prototypical and more prone to distant supervision errors. The patterns scored high by the proposed combination generalize better, variation at the top is achieved by re-combining elements that carry relational meaning (\"is an\", \"vice president\", \"president director\") or are closely correlated to the particular relation. a feature-based extension of a hierarchical topic model, and an at-least-one perceptron. Interpolation increases the quality of extractions and achieves state-of-the-art extraction performance. A combination scheme based on non-dominated sorting, that was inspired by work on combining machine translation metrics, was not as good as a simple linear combination of scores. We think that the good results motivate research into more integrated combinations of noise reduction approaches.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Illustration: Top-Ranked Patterns",
"sec_num": "4.3"
},
{
"text": "The hyper-parameters used for the feature-based topic model are \u03b1 = (1, 1, 1) and \u03b2 = (.1, .001, .001).",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
},
{
"text": "A data point h1 dominates a data point h2 if h1 \u2265 h2 in all metrics and h1 > h2 in at least one metric.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "",
"sec_num": null
}
],
"back_matter": [
{
"text": "Benjamin Roth is a recipient of the Google Europe Fellowship in Natural Language Processing, and this research is supported in part by this Google Fellowship.",
"cite_spans": [],
"ref_spans": [],
"eq_spans": [],
"section": "Acknowledgment",
"sec_num": null
}
],
"bib_entries": {
"BIBREF4": {
"ref_id": "b4",
"title": "s responsibility to pin down just how the government decided to front $ 30 billion in taxpayer dollars for the Bear Stearns deal",
"authors": [],
"year": null,
"venue": "",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "*[ARG1] 's responsibility to pin down just how the government decided to front $ 30 billion in taxpayer dollars for the Bear Stearns deal , \" Chairman [ARG2] org:top members employees, perc+hier (itpl) [ARG2] , Vice President of the [ARG1] [ARG1] Vice president [ARG2] [ARG1] president director [ARG2] [ARG1] vice president director [ARG2] [ARG1] Board member [ARG2]",
"links": null
},
"BIBREF5": {
"ref_id": "b5",
"title": "Pattern learning for relation extraction with a hierarchical topic model",
"authors": [
{
"first": "Enrique",
"middle": [],
"last": "Alfonseca",
"suffix": ""
},
{
"first": "Katja",
"middle": [],
"last": "Filippova",
"suffix": ""
},
{
"first": "Jean-Yves",
"middle": [],
"last": "Delort",
"suffix": ""
},
{
"first": "Guillermo",
"middle": [],
"last": "Garrido",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Short Papers",
"volume": "2",
"issue": "",
"pages": "54--59",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Enrique Alfonseca, Katja Filippova, Jean-Yves Delort, and Guillermo Garrido. 2012. Pattern learning for relation extraction with a hierarchical topic model. In Proceedings of the 50th Annual Meeting of the Asso- ciation for Computational Linguistics: Short Papers- Volume 2, pages 54-59. Association for Computa- tional Linguistics.",
"links": null
},
"BIBREF6": {
"ref_id": "b6",
"title": "The skyline operator",
"authors": [
{
"first": "S",
"middle": [],
"last": "Borzsony",
"suffix": ""
},
{
"first": "Donald",
"middle": [],
"last": "Kossmann",
"suffix": ""
},
{
"first": "Konrad",
"middle": [],
"last": "Stocker",
"suffix": ""
}
],
"year": 2001,
"venue": "Proceedings. 17th International Conference on",
"volume": "",
"issue": "",
"pages": "421--430",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "S Borzsony, Donald Kossmann, and Konrad Stocker. 2001. The skyline operator. In Data Engineering, 2001. Proceedings. 17th International Conference on, pages 421-430. IEEE.",
"links": null
},
"BIBREF7": {
"ref_id": "b7",
"title": "Learning to translate with multiple objectives",
"authors": [
{
"first": "Kevin",
"middle": [],
"last": "Duh",
"suffix": ""
},
{
"first": "Katsuhito",
"middle": [],
"last": "Sudoh",
"suffix": ""
},
{
"first": "Xianchao",
"middle": [],
"last": "Wu",
"suffix": ""
},
{
"first": "Hajime",
"middle": [],
"last": "Tsukada",
"suffix": ""
},
{
"first": "Masaaki",
"middle": [],
"last": "Nagata",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers",
"volume": "1",
"issue": "",
"pages": "1--10",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Kevin Duh, Katsuhito Sudoh, Xianchao Wu, Hajime Tsukada, and Masaaki Nagata. 2012. Learning to translate with multiple objectives. In Proceedings of the 50th Annual Meeting of the Association for Com- putational Linguistics: Long Papers-Volume 1, pages 1-10. Association for Computational Linguistics.",
"links": null
},
"BIBREF8": {
"ref_id": "b8",
"title": "Algorithms and analyses for maximal vector computation",
"authors": [
{
"first": "Parke",
"middle": [],
"last": "Godfrey",
"suffix": ""
},
{
"first": "Ryan",
"middle": [],
"last": "Shipley",
"suffix": ""
},
{
"first": "Jarek",
"middle": [],
"last": "Gryz",
"suffix": ""
}
],
"year": 2007,
"venue": "The VLDB JournalThe International Journal on Very Large Data Bases",
"volume": "16",
"issue": "",
"pages": "5--28",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Parke Godfrey, Ryan Shipley, and Jarek Gryz. 2007. Al- gorithms and analyses for maximal vector computa- tion. The VLDB JournalThe International Journal on Very Large Data Bases, 16(1):5-28.",
"links": null
},
"BIBREF9": {
"ref_id": "b9",
"title": "Knowledgebased weak supervision for information extraction of overlapping relations",
"authors": [
{
"first": "Raphael",
"middle": [],
"last": "Hoffmann",
"suffix": ""
},
{
"first": "Congle",
"middle": [],
"last": "Zhang",
"suffix": ""
},
{
"first": "Xiao",
"middle": [],
"last": "Ling",
"suffix": ""
},
{
"first": "Luke",
"middle": [],
"last": "Zettlemoyer",
"suffix": ""
},
{
"first": "Daniel",
"middle": [
"S"
],
"last": "Weld",
"suffix": ""
}
],
"year": 2011,
"venue": "Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies",
"volume": "",
"issue": "",
"pages": "541--550",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Raphael Hoffmann, Congle Zhang, Xiao Ling, Luke Zettlemoyer, and Daniel S Weld. 2011. Knowledge- based weak supervision for information extraction of overlapping relations. In Proceedings of the 49th An- nual Meeting of the Association for Computational Linguistics: Human Language Technologies, vol- ume 1, pages 541-550.",
"links": null
},
"BIBREF10": {
"ref_id": "b10",
"title": "Overview of the tac 2010 knowledge base population track",
"authors": [
{
"first": "Heng",
"middle": [],
"last": "Ji",
"suffix": ""
},
{
"first": "Ralph",
"middle": [],
"last": "Grishman",
"suffix": ""
},
{
"first": "Hoa",
"middle": [
"Trang"
],
"last": "Dang",
"suffix": ""
},
{
"first": "Kira",
"middle": [],
"last": "Griffitt",
"suffix": ""
},
{
"first": "Joe",
"middle": [
"Ellis"
],
"last": "",
"suffix": ""
}
],
"year": 2010,
"venue": "Third Text Analysis Conference",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Heng Ji, Ralph Grishman, Hoa Trang Dang, Kira Grif- fitt, and Joe Ellis. 2010. Overview of the tac 2010 knowledge base population track. In Third Text Anal- ysis Conference (TAC 2010).",
"links": null
},
"BIBREF11": {
"ref_id": "b11",
"title": "Distant supervision for relation extraction without labeled data",
"authors": [
{
"first": "Mike",
"middle": [],
"last": "Mintz",
"suffix": ""
},
{
"first": "Steven",
"middle": [],
"last": "Bills",
"suffix": ""
},
{
"first": "Rion",
"middle": [],
"last": "Snow",
"suffix": ""
},
{
"first": "Dan",
"middle": [],
"last": "Jurafsky",
"suffix": ""
}
],
"year": 2009,
"venue": "Proceedings of the Joint Conference of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Language Processing of the AFNLP",
"volume": "2",
"issue": "",
"pages": "1003--1011",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mike Mintz, Steven Bills, Rion Snow, and Dan Jurafsky. 2009. Distant supervision for relation extraction with- out labeled data. In Proceedings of the Joint Confer- ence of the 47th Annual Meeting of the ACL and the 4th International Joint Conference on Natural Lan- guage Processing of the AFNLP: Volume 2-Volume 2, pages 1003-1011. Association for Computational Lin- guistics.",
"links": null
},
"BIBREF12": {
"ref_id": "b12",
"title": "Modeling relations and their mentions without labeled text",
"authors": [
{
"first": "Sebastian",
"middle": [],
"last": "Riedel",
"suffix": ""
},
{
"first": "Limin",
"middle": [],
"last": "Yao",
"suffix": ""
},
{
"first": "Andrew",
"middle": [],
"last": "Mccallum",
"suffix": ""
}
],
"year": 2010,
"venue": "Machine Learning and Knowledge Discovery in Databases",
"volume": "",
"issue": "",
"pages": "148--163",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Sebastian Riedel, Limin Yao, and Andrew McCallum. 2010. Modeling relations and their mentions with- out labeled text. In Machine Learning and Knowledge Discovery in Databases, pages 148-163. Springer.",
"links": null
},
"BIBREF13": {
"ref_id": "b13",
"title": "Featurebased models for improving the quality of noisy training data for relation extraction",
"authors": [
{
"first": "Benjamin",
"middle": [],
"last": "Roth",
"suffix": ""
},
{
"first": "Dietrich",
"middle": [],
"last": "Klakow",
"suffix": ""
}
],
"year": 2013,
"venue": "Proceedings of the 22nd ACM International Conference on Information and Knowledge Management (CIKM)",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Benjamin Roth and Dietrich Klakow. 2013. Feature- based models for improving the quality of noisy train- ing data for relation extraction. In Proceedings of the 22nd ACM International Conference on Information and Knowledge Management (CIKM). ACM.",
"links": null
},
"BIBREF14": {
"ref_id": "b14",
"title": "Generalizing from freebase and patterns using distant supervision for slot filling",
"authors": [
{
"first": "Benjamin",
"middle": [],
"last": "Roth",
"suffix": ""
},
{
"first": "Grzegorz",
"middle": [],
"last": "Chrupala",
"suffix": ""
},
{
"first": "Michael",
"middle": [],
"last": "Wiegand",
"suffix": ""
},
{
"first": "Mittul",
"middle": [],
"last": "Singh",
"suffix": ""
},
{
"first": "Dietrich",
"middle": [],
"last": "Klakow",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the Text Analysis Conference",
"volume": "",
"issue": "",
"pages": "",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Benjamin Roth, Grzegorz Chrupala, Michael Wiegand, Mittul Singh, and Dietrich Klakow. 2012. General- izing from freebase and patterns using distant supervi- sion for slot filling. In Proceedings of the Text Analysis Conference (TAC).",
"links": null
},
"BIBREF15": {
"ref_id": "b15",
"title": "Multi-instance multilabel learning for relation extraction",
"authors": [
{
"first": "Mihai",
"middle": [],
"last": "Surdeanu",
"suffix": ""
},
{
"first": "Julie",
"middle": [],
"last": "Tibshirani",
"suffix": ""
},
{
"first": "Ramesh",
"middle": [],
"last": "Nallapati",
"suffix": ""
},
{
"first": "Christopher D",
"middle": [],
"last": "Manning",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning",
"volume": "",
"issue": "",
"pages": "455--465",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Mihai Surdeanu, Julie Tibshirani, Ramesh Nallapati, and Christopher D Manning. 2012. Multi-instance multi- label learning for relation extraction. In Proceedings of the 2012 Joint Conference on Empirical Methods in Natural Language Processing and Computational Natural Language Learning, pages 455-465. Associa- tion for Computational Linguistics.",
"links": null
},
"BIBREF16": {
"ref_id": "b16",
"title": "Reducing wrong labels in distant supervision for relation extraction",
"authors": [
{
"first": "Shingo",
"middle": [],
"last": "Takamatsu",
"suffix": ""
},
{
"first": "Issei",
"middle": [],
"last": "Sato",
"suffix": ""
},
{
"first": "Hiroshi",
"middle": [],
"last": "Nakagawa",
"suffix": ""
}
],
"year": 2012,
"venue": "Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers",
"volume": "1",
"issue": "",
"pages": "721--729",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Shingo Takamatsu, Issei Sato, and Hiroshi Nakagawa. 2012. Reducing wrong labels in distant supervi- sion for relation extraction. In Proceedings of the 50th Annual Meeting of the Association for Compu- tational Linguistics: Long Papers -Volume 1, ACL '12, pages 721-729, Stroudsburg, PA, USA. Associ- ation for Computational Linguistics.",
"links": null
},
"BIBREF17": {
"ref_id": "b17",
"title": "Collective cross-document relation extraction without labelled data",
"authors": [
{
"first": "Limin",
"middle": [],
"last": "Yao",
"suffix": ""
},
{
"first": "Sebastian",
"middle": [],
"last": "Riedel",
"suffix": ""
},
{
"first": "Andrew",
"middle": [],
"last": "Mccallum",
"suffix": ""
}
],
"year": 2010,
"venue": "Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing",
"volume": "",
"issue": "",
"pages": "1013--1023",
"other_ids": {},
"num": null,
"urls": [],
"raw_text": "Limin Yao, Sebastian Riedel, and Andrew McCallum. 2010. Collective cross-document relation extraction without labelled data. In Proceedings of the 2010 Con- ference on Empirical Methods in Natural Language Processing, pages 1013-1023. Association for Com- putational Linguistics.",
"links": null
}
},
"ref_entries": {
"FIGREF0": {
"text": "Hierarchical topic models. Intertext model (left) and feature model (right).",
"type_str": "figure",
"num": null,
"uris": null
},
"FIGREF1": {
"text": "The weight vectors are averaged over 20 iterations.",
"type_str": "figure",
"num": null,
"uris": null
},
"FIGREF2": {
"text": "Score combination by non-dominated sorting: Circles indicate patterns on the Pareto-frontier, which are ranked highest. They are followed by the triangles, the square indicates the lowest ranked pattern in this example.For the patterns of the form[ARG1] context [ARG2], we compute the following scores:\u2022 Maximum Likelihood (MLE): n(pat,r)\u2022P (r|s,\u03b8) n(pat)\u2022(P (r|s,\u03b8)+P (N IL|s,\u03b8))",
"type_str": "figure",
"num": null,
"uris": null
},
"FIGREF3": {
"text": "Precision at recall levels.",
"type_str": "figure",
"num": null,
"uris": null
},
"FIGREF4": {
"text": "Top-scored patterns for maximum likelihood (MLE) and the interpolation (perc+hier itpl) method. Inexact patterns are marked by *.",
"type_str": "figure",
"num": null,
"uris": null
}
}
}
} |