ie_introduction.tex 27.7 KB
Newer Older
Loïc Barrault's avatar
Loïc Barrault committed
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's avatar
Loïc Barrault committed
24
	\item \textbf{Brief(est) history of IE}
Loïc Barrault's avatar
Loïc Barrault committed
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
\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

Loïc Barrault's avatar
ie    
Loïc Barrault committed
93
%\todo{ADD A GRAPHIC WITH OBAMA, events = president start, end... WIFE = michelle etc...}
Loïc Barrault's avatar
Loïc Barrault committed
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{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's avatar
Loïc Barrault committed
221
	\item Brief(est) history of IE
Loïc Barrault's avatar
Loïc Barrault committed
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
\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
Loïc Barrault's avatar
ie    
Loïc Barrault committed
332
\item \textbf{Relation classification}: determine the type of a previously detected relation
Loïc Barrault's avatar
Loïc Barrault committed
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
\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}

Loïc Barrault's avatar
ie    
Loïc Barrault committed
387
\begin{block}{\textbf{Event Extraction}}
Loïc Barrault's avatar
Loïc Barrault committed
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
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's avatar
Loïc Barrault committed
441
	\item Briefest history of IE
Loïc Barrault's avatar
Loïc Barrault committed
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
\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's avatar
Loïc Barrault committed
615
	\item Briefest history of IE
Loïc Barrault's avatar
Loïc Barrault committed
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
\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's avatar
Loïc Barrault committed
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
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\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's avatar
Loïc Barrault committed
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
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\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}