ie_introduction.tex 27.7 KB
Newer Older
 Loïc Barrault committed Nov 10, 2019 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 % !TEX root = text_processing.tex %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{} \vfill \centering \Huge{\edinred{[Information Extraction]\\Introduction}} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{} \begin{itemize} \item \textbf{Introduction to Information Extraction} \begin{itemize} \item \textbf{Definition + contrast with IR} \item \textbf{Example Applications} \item \textbf{Overview of Tasks} \item \textbf{Overview of Approaches} \item \textbf{Evaluation + Shared Task Challenges}  Loïc Barrault committed Nov 10, 2019 24  \item \textbf{Brief(est) history of IE}  Loïc Barrault committed Nov 10, 2019 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 \end{itemize} \item Named Entity Recognition \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning \item Entity Linking \end{itemize} \item Relation Extraction \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning, Bootstrapping, Distant Supervision \end{itemize} \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction} \begin{block}{Definition} \myemph{Information Extraction} is the task of \textbf{identifying information} about predefined classes of \textbf{entitites}, \textbf{relationships} or \textbf{events} and record it in a \textbf{structured form} \end{block} \vfill \textbf{Other definitions:} \begin{itemize} \item The activity of populating a structured information repository from an unstructured information source \item The activity of creating a semantically annotated text collection ~(\ra\ \myemph{semantic web}) \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction} Why doing \myemph{information extraction}? \begin{itemize} \item searching or analysis using conventional database queries \begin{itemize} \item[\ra] Difficult to search directly in the text because it lacks structure. \end{itemize} \item data mining \item summarisation (eventually in another language) \item construct indexes into/within/between large quantity of text \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction} \begin{itemize} \item Entities: persons, organisations, locations, times, etc. \ra\ \myemph{Named Entities} \item Relationships: links between entities, etc. \item Events: succession events, etc. \end{itemize} \vfill The \textbf{structured form} can be implemented either as a database or form (slot filling) or by using XML tags (tagging) \vfill \todo{ADD A GRAPHIC WITH OBAMA, events = president start, end... WIFE = michelle etc...} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Example} {\centering \fbox{ \parbox{.9\textwidth}{ Who’s News: @ \myhl<3->{cyan!40}{Burns Fry Ltd.}\\ \myhl<5->{orange}{04/13/94} WALL STREET JOURNAL (J), PAGE B10 \textlangle CO\textrangle\\ \myhl<3->{cyan!40}{BURNS FRY Ltd.} (\myhl<4->{green!30}{Toronto}) – \myhl<2->{brown!90}{Donald Wright}, 46 years old, \myhl<7->{blue!20}{was named} \myhl<6->{red!40}{executive vice president} and \myhl<6->{red!40}{director of fixed income} at this brokerage firm. \myhl<2->{brown!90}{Mr. Wright} \myhl<7->{blue!20}{resigned} as \myhl<6->{red!40}{president} of \myhl<3->{cyan!40}{Merrill Lynch Canada Inc.}, a unit of \myhl<3->{cyan!40}{Merrill Lynch \& Co.}, to succeed \myhl<2->{brown!90}{Mark Kassirer}, 48, who left \myhl<3->{cyan!40}{Burns Fry} \myhl<5->{orange}{last month}. A \myhl<3->{cyan!40}{Merrill Lynch} \myhl<6->{red!40}{spokeswoman} said it hasn’t named a successor to \myhl<2->{brown!90}{Mr. Wright}, who is expected to begin his new position \myhl<5->{orange}{by the end of the month}. }} } \begin{columns} \begin{column}{.5\textwidth} \begin{itemize} \item \myhl<2->{brown!90}{persons} \item \myhl<3->{cyan!40}{organisations} \item \myhl<4->{green!30}{locations} \end{itemize} \end{column} \begin{column}{.5\textwidth} \begin{itemize} \item \myhl<5->{orange}{times} \item \myhl<6->{red!40}{position in a company} \item \myhl<7->{blue!20}{succession events} \end{itemize} \end{column} \end{columns} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Filled template} \begin{center} \includegraphics[width=0.8\textwidth]{ie_template} \end{center} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction vs. Information Retrieval} \textbf{IR Task:} \begin{itemize} \item Given a document collection and a user query \item Returns a (ranked) list of documents relevant to the user query \end{itemize} \textbf{Strengths:} \begin{itemize} \item Can search huge document collections very rapidly \item Insensitive to genre and domain of the texts \item Relatively straightforward to implement \begin{itemize} \item challenges scaling to huge, dynamic document collections, e.g. the web \end{itemize} \end{itemize} \textbf{Weaknesses} \begin{itemize} \item Documents are returned rather than information/answers \begin{itemize} \item user must further read texts to extract information \item output is unstructured so limited possibilities for further processing \end{itemize} \item Frequently not discriminating enough (“14,100,000 documents match your request”) \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction vs. Information Retrieval} \textbf{IE Task:} \begin{itemize} \item Given a document collection and a predefined set of entities, relations and/or events \item Returns a structured representation of all mentions of the specified entities, relations and/or events \end{itemize} \textbf{Strengths:} \begin{itemize} \item Extracts \textbf{facts} from texts, not just texts from text collections \item Can feed other powerful applications (databases, semantic indexing engines, data mining tools) \end{itemize} \textbf{Weaknesses:} \begin{itemize} \item Systems tend to be genre/domain specific and porting to new genres and domains can be time-consuming/requires expertise \item Limited accuracy \item Computationally demanding, so performance issues on very large collections \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: example applications} \begin{itemize} \item Google News uses Named Entity Recognition for its “In the News” feature \item Scrapping web pages to build structured databases of job postings, apartment rentals, seminar announcements, etc. \item Assisting biomedical database curators by extracting biomedical entities and relations from the scientific literature prior to entry in a human-maintained database (e.g. Flybase) \item Assisting companies in competitor intelligence gathering, e.g. management or researcher succession events, new product or project annoucements, etc. \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Overview} \begin{itemize} \item \textbf{Introduction to Information Extraction} \begin{itemize} \item \gray{Definition + contrast with IR} \item \gray{Example Applications} \item \textbf{Overview of Tasks} \item Overview of Approaches \item Evaluation + Shared Task Challenges  Loïc Barrault committed Nov 10, 2019 221  \item Brief(est) history of IE  Loïc Barrault committed Nov 10, 2019 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 \end{itemize} \item Named Entity Recognition \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning \item Entity Linking \end{itemize} \item Relation Extraction \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning, Bootstrapping, Distant Supervision \end{itemize} \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Overview of tasks} \begin{block}{\textbf{Entity Extraction/Named Entity Recognition (NER)}} Task: Identify the \myemph{extent} and the \myemph{type} of each textual mention of an entity\\ The set of types is determined in advance (e.g. organisation, person, date, etc...) \end{block} \begin{center} \begin{tabular}{ll} \myhl{cyan!40}{Cable and Wireless} today announced \ldots & Extent: 0-3 ; Type = \myhl{cyan!40}{ORG} \\ \myhl{cyan!40}{IBM} and \myhl{cyan!40}{Microsoft} today announced \ldots & Extent: 0-1 ; Type = \myhl{cyan!40}{ORG} \\ & Extent: 2-3 ; Type = \myhl{cyan!40}{ORG} \\ \myhl{brown!90}{John Lewis} hired \ldots & Extent: 0-2 ; Type = \myhl{cyan!40}{ORG} \\ \myhl{brown!90}{Theresa May} hired. & Extent: 0-2 ; Type = \myhl{brown!90}{PER} \end{tabular} \end{center} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Entity extraction} \textbf{Types of entities addressed by IE systems include:}\\ \begin{itemize} \item \textbf{Named individuals} \begin{itemize} \item Organisations (ORG), persons (PER), books, films, ships, restaurants . . . \item[\ra] \myhl{cyan!40}{Cable and Wireless} today announced \ldots ; Extent: \textbf{0-3} ; Type = \textbf{ORG} \\ \item[\ra] \myhl{brown!90}{Barack Obama} was the 44th president... \ldots ; Extent: \textbf{0-3} ; Type = \textbf{PER} \\ \item Geo-Political entities (GPE), locations (LOC) \item[\ra] The \myhl{carminered}{Mont Blanc} intersects France, Italy and Switzerland. ; Extent: \textbf{1-3} ; Type = \textbf{LOC} \\ \item[\ra] The Mont Blanc intersects \myhl{carminered!60}{France}, \myhl{carminered!60}{Italy} and \myhl{carminered!60}{Switzerland}. ; Extent: \textbf{4-5} ; Type = \textbf{GPE} \\ \end{itemize} %\item Named kinds %\begin{itemize} %\item Proteins, chemical compounds/drugs, diseases, aircraft components . . . %\end{itemize} \item \textbf{Times}: temporal expressions dates, times of day \begin{itemize} \item[\ra] Let's meet at \myhl{orange}{2pm} next Friday \ldots ; Extent: \textbf{3-4} ; Type = \textbf{TIME} \\ \item[\ra] Let's meet at 2pm next \myhl{orange!50}{Friday} \ldots ; Extent: \textbf{5-6} ; Type = \textbf{DATE} \\ \end{itemize} \item \textbf{Measures}: monetary expressions, distances/sizes, weights . . . \begin{itemize} \item[\ra] This watch costs \myhl{bananayellow}{£35} \ldots ; Extent: \textbf{3-4} ; Type = \textbf{MONEY} \\ \end{itemize} \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Entity extraction: coreference} \begin{block}{\textbf{Coreference}} Different textual expressions that refer to the same real world entity are said to \myemph{corefer}. \textbf{Coreference Task}: link together all textual references to the same \myemph{real world entity}, \end{block} Multiple references to the same entity in a text are rarely made using the same string: \begin{itemize} \item Pronouns: \textbf{Tony Blair} == \textbf{he} \item Names/definite descriptions: \textbf{Tony Blair} == \textbf{the Prime Minister} \item Abbreviated forms: \textbf{Theresa May} == \textbf{May}; \textbf{European Union} == \textbf{EU} \item Orthographic variants: \textbf{alpha helix} == \textbf{alpha-helix} == \textbf{$\bm{\alpha}$-helix} == \textbf{a-helix} \end{itemize} \vfill Can be seen as a separate task or as part of entity extraction task \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Relation extraction} \begin{block}{\textbf{Relation Extraction}} Identify all assertions of relations between entities %usually binary \end{block} May be divided into two subtasks: \begin{itemize} \item \textbf{Relation detection}: find pairs of entities between which a relation holds \item \textbf{Relation classification}: determine the type of a previously extracted relation \end{itemize} \vfill \only<2>{ Example: {\sc location\_of} holding between \begin{itemize} \item[] \begin{itemize} \item {\sc organisation} and {\sc geopolotical\_location} \item medical {\sc investigation} and {\sc body\_part} \item {\sc gene} and {\sc chromosome\_location} \end{itemize} \item {\sc employee\-of} holding between {\sc person} and {\sc organisation} \item {\sc product\_of} holding between {\sc artifact} and {\sc organisation} %\item {\sc interaction} holding between {\sc protein} and {\sc protein} \end{itemize} } \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Relation extraction} \textbf{Challenges for Relation Extraction} \begin{enumerate} \item<1-> Many different ways to express the same relation \begin{itemize} \item \myemph{Canonical}: \annot{Microsoft}{ORG} \textbf{\underline{is located in}} \annot{Redmond}{LOC} \item \myemph{Synonyms}: \annot{Microsoft}{ORG} \textbf{\underline{is located/based/headquartered in}} \annot{Redmond}{LOC} \item \myemph{Syntactic variations}: \begin{itemize} \item \annot{Microsoft}{ORG}, the software giant and ... , \textbf{\underline{is based in}} \annot{Redmond}{LOC} \item \annot{Redmond}{LOC}\textbf{\underline{-based}} \annot{Microsoft}{ORG} ... \item \annot{Redmond}{LOC}\textbf{\underline{'s}} \annot{Microsoft}{ORG} ... \item \annot{Redmond}{LOC} software giant \annot{Microsoft}{ORG} ... \end{itemize} \end{itemize} \item<2-> Relations often involve coreference links\\ \small{ \myemph{\annot{Bill Gates}{PER}} \textbf{\underline{is co-founder, technology advisor and board member}} of \annot{Microsoft}{ORG}. \myemph{\annot{He}{PER}} \textbf{\underline{served as chairman of the board}} until Feb. 4, 2014. On June 27, 2008, \myemph{\annot{Gates}{PER}} \textbf{\underline{transitioned out}} of a day-to-day role in the company to spend more time on his global health and education work at the \annot{Bill \& Melinda Gates Foundation}{ORG}. } \end{enumerate} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Event extraction} \begin{block}{\textbf{Relation Extraction}} Identify all reports of event instances, typically of a small set of classes \end{block} May be divided into two subtasks: \begin{itemize} \item \textbf{Event detection}: find all mentions of events in a text \item \textbf{Event classification}: assign a class to the detected events \end{itemize} \vfill Examples \begin{itemize} \item National/european elections \item Management succession events \item Joint venture/product announcements \item Terrorist attacks \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Event extraction} \textbf{Challenges for Event Extraction}\\ Events may be simply viewed as relations. However they are typically complex relations\\ \vfill \begin{itemize} \item<1-> often temporally situated, often short duration \item[\ra]<1-> \small{Bolt etched his name in history with a 9.69-second finish \textbf{on Aug. 16, 2008}.}\\ \item[]<2-> \item<2-> often involve multiple role players (often >2) \item[\ra]<2-> \small{23 September 2019. \textbf{\underline{Banks, businesses, civil society and governments}} at all levels are to announce initiatives to finance and build a new generation of sustainable cities at the \textbf{UN Climate Action Summit} in New York today.}\\ \item[]<3-> \item<3-> often expressed across multiple sentences \end{itemize} \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Overview} \begin{itemize} \item \textbf{Introduction to Information Extraction} \begin{itemize} \item \gray{Definition + contrast with IR} \item \gray{Example Applications} \item \gray{Overview of Tasks} \item \textbf{Overview of Approaches} \item Evaluation + Shared Task Challenges  Loïc Barrault committed Nov 10, 2019 442  \item Briefest history of IE  Loïc Barrault committed Nov 10, 2019 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 \end{itemize} \item Named Entity Recognition \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning \item Entity Linking \end{itemize} \item Relation Extraction \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning, Bootstrapping, Distant Supervision \end{itemize} \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Overview of Approaches} \vfill Approaches to IE may be placed into four categories: \begin{enumerate} \item Knowledge Engineering Approaches \item Supervised Learning Approaches \item Bootstrapping Approaches \item Distant Supervision Approaches \end{enumerate} \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Knowledge Engineering Approaches} \colorbox{liumlightgray}{ \parbox{.99\textwidth}{ \myhl{brown!90}{\annot{Mr. \tikzmark{a} Wright}{PER}}, \textbf{\underline{\annot{executive vice president}{POSITION}}} of \myhl{cyan!40}{\annot{Merrill Lynch \tikzmark{b} Canada Inc.}{ORG}}\\ \begin{tikzpicture}[overlay,remember picture] \draw [very thick, color=carminered] ($({pic cs:a})+(0ex,-1ex)$) -- ($({pic cs:a})+(0ex,-3ex)$); \draw [very thick, color=carminered] ($({pic cs:a})+(0ex,-3ex)$) -- ($({pic cs:b})+(0ex,-3ex)$) node [midway, below, color=carminered] {is-employed-by}; \draw [very thick, color=carminered] ($({pic cs:b})+(0ex,-3ex)$) -- ($({pic cs:b})+(0ex,-1ex)$); \end{tikzpicture} }} \vfill Use manually authored rules and can be divided into \begin{itemize} \item “deep” – linguistically inspired language understanding systems \item “shallow” – systems engineered to the IE task, typically using pattern-action rules \end{itemize} \begin{center} \begin{tabular}{ll} Pattern: & ‘\textbf{‘Mr. \$Uppercase-initial-word’’} \\ Action: & \textbf{add-entity(person(Mr. \$Uppercase-initial-word))} \\ & \\ Pattern: & \textbf{"\$Person, \$Position of \$Organization"} \\ Action: & \textbf{add-relation(is-employed-by(\$Person,\$Organization))} \\ \end{tabular} \end{center} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Supervised Learning Approaches} \Ra\ Machine Learning systems trained on manually annotated texts (entities and relations)\\ \colorbox{liumlightgray}{ \parbox{.99\textwidth}{ \myhl{brown!90}{\annot{Mr. \tikzmark{a2} Wright}{PER}}, \textbf{\underline{\annot{executive vice president}{POSITION}}} of \myhl{cyan!40}{\annot{Merrill Lynch \tikzmark{b2} Canada Inc.}{ORG}}\\ \begin{tikzpicture}[overlay,remember picture] \draw [very thick, color=carminered] ($({pic cs:a2})+(0ex,-1ex)$) -- ($({pic cs:a2})+(0ex,-3ex)$); \draw [very thick, color=carminered] ($({pic cs:a2})+(0ex,-3ex)$) -- ($({pic cs:b2})+(0ex,-3ex)$) node [midway, below, color=carminered] {is-employed-by}; \draw [very thick, color=carminered] ($({pic cs:b2})+(0ex,-3ex)$) -- ($({pic cs:b2})+(0ex,-1ex)$); \end{tikzpicture} }} \textbf{For each entity/relation create a training instance} \begin{itemize} \item$k$words either side of an entity mention \item$k$words to the left of entity 1 and to the right of entity 2 plus the words in between \item[\ra] extract \myemph{features}: words, POS, morphology \end{itemize} \textbf{Systems may learn} \begin{itemize} \item patterns that match extraction targets \item classifiers that classify tokens as beginning/inside/outside a tag type \end{itemize} \textbf{Machine Learning techniques}: covering algorithms, HMMs, SVMs \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Bootstrapping Approaches} \Ra\ A technique for relation extraction that requires only minimal supervision \vfill Systems are given \begin{itemize} \item seed tuples: e.g. \myemph{\textlangle Microsoft , Redmond\textrangle} \item seed patterns: e.g. \myemph{\annot{X}{ORG}} \textbf{\underline{is located in}} \myemph{\annot{Y}{LOC}} \end{itemize} \vfill System searches in large corpus for: \begin{itemize} %\item occurrences of seed tuples and then extracts a \item patterns that matches the context of a known seed tuple \ra\ add them to the \textbf{seed patterns} \item new tuples that match a known seed patterns \ra\ add them to the \textbf{seed tuples} \item[\ra] process iterates until convergence \item[\ra] detailed in a later lecture \end{itemize} \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Distant Supervision Approaches} \Ra\ also called “weakly labelled” or "lightly supervised" approaches \begin{enumerate} \item Assumes a (semi-)structured data source, such as \begin{itemize} \item Wikipedia infoboxes (e.g. \myemph{\sc{person born\_in location/date}}) %\item Freebase, Wikidata \item Yeast Protein Database, (e.g. \myemph{{\sc protein is\_located\_in subcellular\_ location}}) \item[\ra] contains tuples of entities standing in the relation of interest \item[\ra] eventually pointing to the source text \end{itemize} \item Tuples from data source are used to label \begin{itemize} \item the text with which they are associated, if available \item documents from the web, if not \end{itemize} \item \textbf{Labelled data is used to train a standard supervised NER or RE system}\\ \ra\ See later lecture \end{enumerate} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Overview} \begin{itemize} \item \textbf{Introduction to Information Extraction} \begin{itemize} \item \gray{Definition + contrast with IR} \item \gray{Example Applications} \item \gray{Overview of Tasks} \item \gray{Overview of Approaches} \item \textbf{Evaluation + Shared Task Challenges}  Loïc Barrault committed Nov 10, 2019 616  \item Briefest history of IE  Loïc Barrault committed Nov 10, 2019 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 \end{itemize} \item Named Entity Recognition \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning \item Entity Linking \end{itemize} \item Relation Extraction \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning, Bootstrapping, Distant Supervision \end{itemize} \end{itemize} \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Evaluation} \textbf{How to evaluate an IE system?} \vfill Needs a fully annotated corpus (\textbf{test}) \ra\ produced manually \begin{itemize} \item Correct answers, called \myemph{keys}, or \textbf{reference} or \textbf{ground truth} \item ideally: several annotations by several annotators \ra\ \myemph{interannotator agreement} \end{itemize} \vfill Scoring of system results, called \myemph{responses}, or \myemph{hypotheses} against \textbf{keys} is done automatically. \vfill Principal metrics are: \begin{itemize} \item \myemph{Precision} (how much of what system returns is correct) \item \myemph{Recall} (how much of what is correct system returns) \item \myemph{F-measure} (a weighted combination of precision and recall) \end{itemize} \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Evaluation Metrics} \textbf{Confusion matrix} for a \myemph{binary classification} task \ra\ \begin{columns} \begin{column}{.59\textwidth} \textbf{Precision}: how much of what system returns is correct \begin{equation*} precision = P = \frac{TP}{TP+FP} \end{equation*} \textbf{Recall}: how much of what is correct system returns \begin{equation*} recall = R = \frac{TP}{TP+FN} \end{equation*} \end{column} \begin{column}{.4\textwidth} \includegraphics[width=0.9\textwidth]{confusion_matrix_bin} \end{column} \end{columns} \vfill \textbf{F-measure}: a weighted combination of precision and recall \begin{equation*} F_\beta = \frac{(1+\beta^2)* P * R}{\beta^2 * P + R} \Rightarrow F1 = \frac{2 * P * R}{P + R} \end{equation*} \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Evaluation Metrics} \textbf{Confusion matrix} for a \myemph{multiclass classification} task \ra\ \begin{columns} \begin{column}{.59\textwidth} \textbf{Precision} and \textbf{Recall} for class$i$: \begin{equation*} precision_i = P_i = \frac{R_i\_H_i}{H_i} \text{~and~} recall_i = R_i = \frac{R_i\_H_i}{R_i} \end{equation*} \textbf{Micro} and \textbf{macro} averages \begin{equation*} \text{Micro-}prec = \frac{\sum_i R_i\_H_i}{\sum_i H_i} \text{~and~} \text{Micro-}recall = \frac{\sum_i R_i\_H_i}{\sum_i R_i} \end{equation*} \begin{equation*} \text{Macro-}mes = \frac{\ds \sum_{i} mes_i}{\# classes} \text{~~with$mes \in {prec, recall}\$ } \end{equation*} \end{column} \begin{column}{.4\textwidth} \includegraphics[width=0.99\textwidth]{confusion_matrix_multi} \end{column} \end{columns} \vfill \end{frame} %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Shared Task challenges} \begin{block}{Shared Task challenges or \textbf{evaluation campaigns}} competition among (international) research groups to build systems to address a specific task \end{block} Conditions: \begin{itemize} \item clear \textbf{task definition}, annotated \textbf{text resources} for training, development and testing \item agreed \textbf{evaluation metrics} and \textbf{schedule} %(release of resources, system development, evaluation period) \item[\ra] results are usually discussed during a conference \end{itemize} Shared task challenges in IE include: Message Understanding Conferences (\textbf{MUC}, 1985-1998), Automatic Content Extraction (\textbf{ACE}, 1999-2008), Text Analysis Conference (\textbf{TAC}, 2008-2018), \textbf{BioCreative} (molecular biology, 2004-2016) \vfill \textbf{\Ra\ Define the core methodology of the field and have led to significant progress} %MUC: https://en.wikipedia.org/wiki/Message_Understanding_Conference % https://www-nlpir.nist.gov/related_projects/muc/ %ACE: https://en.wikipedia.org/wiki/Automatic_content_extraction % TAC: https://tac.nist.gov/ % BioCreative: https://en.wikipedia.org/wiki/BioCreative \end{frame}  Loïc Barrault committed Nov 10, 2019 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 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Information Extraction: Overview} \begin{itemize} \item \textbf{Introduction to Information Extraction} \begin{itemize} \item \gray{Definition + contrast with IR} \item \gray{Example Applications} \item \gray{Overview of Tasks} \item \gray{Overview of Approaches} \item \gray{Evaluation + Shared Task Challenges} \item \textbf{Brief(est) history of IE} \end{itemize} \item Named Entity Recognition \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning \item Entity Linking \end{itemize} \item Relation Extraction \begin{itemize} \item Task \item Approaches: Rule-based, Supervised Learning, Bootstrapping, Distant Supervision \end{itemize} \end{itemize} \end{frame}  Loïc Barrault committed Nov 10, 2019 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 %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% \begin{frame} \frametitle{Brief(est) history of IE} \begin{description} \item[\myemph{1960s}] First published work on IE (though it was not called IE at the time) \item[\myemph{1970s}] Precursor: the psychologist Roger Schank’s work on scripts and story understanding \item[\myemph{1980s}] Emergence of some commercial systems (financial transactions and newswires) \item[\ra] MUC-1 in 1987 \item[\myemph{1990s}] MUC ran 7 times until 1998 and significantly advanced the field. \item[\ra] Machine Learning approaches to IE began to appear \item[\myemph{2000s}] ACE (1999-2008); succeeded by TAC (2008-present); \begin{itemize} \item[\ra] BioCreative (IE in the biomedical domain) began (2004-present); \item[\ra] work on IE in other languages began (e.g. Spanish, Japanese, Chinese, Arabic) \end{itemize} \item[\myemph{2010s}] TAC [\url{https://tac.nist.gov}] \ra\ \textbf{knowledge base population} track \item[] \item[\Ra] \textbf{Currently: a number of IE systems on the market \& large, ongoing research effort } \end{description} \end{frame}