Extraction as Classification

What is “Information Extraction” Filling slots in a database from sub-segments of text. As a task: October 14, 2002, 4:00 a.m. PT For years, Microsoft Corporation CEO Bill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation. Today, Microsoft claims to "love" the open- source concept, by which software code is made public to encourage improvement and development by outside programmers. Gates himself says Microsoft will gladly disclose its crown jewels--the coveted code behind the Windows operating system--to select customers. "We can be open source. We love the concept of shared source," said Bill Veghte, a Microsoft VP. "That's a super-important shift for us in terms of code access.“ Richard Stallman, founder of the Free Software Foundation, countered saying… NAME TITLE ORGANIZATION Bill Gates CEO Microsoft Bill Veghte VP Microsoft Richard Stallman founder Free Soft.. IE QA End User

What is “Information Extraction” Information Extraction = segmentation + classification + association + clustering As a family of techniques: October 14, 2002, 4:00 a.m. PT For years, Microsoft Corporation CEO Bill Gates railed against the economic philosophy of open-source software with Orwellian fervor, denouncing its communal licensing as a "cancer" that stifled technological innovation. Today, Microsoft claims to "love" the open- source concept, by which software code is made public to encourage improvement and development by outside programmers. Gates himself says Microsoft will gladly disclose its crown jewels--the coveted code behind the Windows operating system--to select customers. "We can be open source. We love the concept of shared source," said Bill Veghte, a Microsoft VP. "That's a super-important shift for us in terms of code access.“ Richard Stallman, founder of the Free Software Foundation, countered saying… Microsoft Corporation CEO Bill Gates Microsoft Gates Microsoft Bill Veghte Microsoft VP Richard Stallman founder Free Software Foundation aka Named Entity Recognition

Landscape of IE Tasks (1/4): Degree of Formatting Text paragraphs without formatting Grammatical sentences and some formatting & links Non-grammatical snippets, rich formatting & links Tables Astro Teller is the CEO and co-founder of BodyMedia. Astro holds a Ph.D. in Artificial Intelligence from Carnegie Mellon University, where he was inducted as a national Hertz fellow. His M.S. in symbolic and heuristic computation and B.S. in computer science are from Stanford University. His work in science, literature and business has appeared in international media from the New York Times to CNN to NPR.

Landscape of IE Tasks (3/4): Complexity of extraction task Closed set He was born in Alabama… Regular set Phone: (413) Complex pattern University of Arkansas P.O. Box 140 Hope, AR …was among the six houses sold by Hope Feldman that year. Ambiguous patterns, needing context and many sources of evidence The CALD main office can be reached at The big Wyoming sky… U.S. states U.S. phone numbers U.S. postal addresses Person names Headquarters: 1128 Main Street, 4th Floor Cincinnati, Ohio Pawel Opalinski, Software Engineer at WhizBang Labs. E.g. word patterns:

Landscape of IE Tasks (4/4): Single Field/Record Single entity Person: Jack Welch Binary relationship Relation: Person-Title Person: Jack Welch Title: CEO N-ary record “Named entity” extraction Jack Welch will retire as CEO of General Electric tomorrow. The top role at the Connecticut company will be filled by Jeffrey Immelt. Relation: Company-Location Company: General Electric Location: Connecticut Relation: Succession Company: General Electric Title: CEO Out: Jack Welsh In: Jeffrey Immelt Person: Jeffrey Immelt Location: Connecticut

Models for NER Lexicons Alabama Alaska … Wisconsin Wyoming Abraham Lincoln was born in Kentucky. member? Classify Pre-segmented Candidates Abraham Lincoln was born in Kentucky. Classifier which class? Sliding Window Abraham Lincoln was born in Kentucky. Classifier which class? Try alternate window sizes: Boundary Models Abraham Lincoln was born in Kentucky. Classifier which class? BEGINENDBEGINEND BEGIN Token Tagging Abraham Lincoln was born in Kentucky. Most likely state sequence? This is often treated as a structured prediction problem…classifying tokens sequentially HMMs, CRFs, ….

MUC-7 Last Message Understanding Conference, (fore-runner to ACE), about 1998 –200 articles in development set (newswire text about aircraft accidents) –200 articles in final test (launch events) –Names of: persons, organizations, locations, dates, times, currency & percentage

TE: Template Elements (attributes) TR: Template Relations (binary relations) ST: Scenario Template (events) CO: coreference NE: named entity recognition

LTG Identifinder (HMMs) MENE+Proteus Manitoba (NB filtered names) NetOwl Commercial RBS

Bothwick, Sterling, Agichtein, Grishman: the MENE system ?

Borthwick et al: MENE system Simple idea: tag every token: –4 tags/field for x=person, organization: x_start, x_continue, x_end, x_unique Also “other” –To extract: Compute P(y|xi) for each xi Use Viterbi to find ML consistent sequence: –continue follows start –end follows continue or start –…

Borthwick et al: MENE system Simple idea: tag every token: –4 tags/field for x=person, organization: x_start, x_continue, x_end, x_unique Also “other” –Learner: Maxent/logistic regression Regularize by dropping rare features –To extract: Compute P(y|x i ) for each x i Use Viterbi to find ML consistent sequence: –continue follows start –end follows continue or start –…

Viterbi in MENE

Borthwick et al: MENE system Features g(h,f) for the loglinear model –function of “history” (features of token) and “future” (predicted class) –Lexical features combine Identity of token in window and Predicted class eg g(h,f)=[token -1 (h)=“Mr.” and f=person_unique] –Section features combine: section name & class

Borthwick et al: MENE system Features g(h,f) for the loglinear model –Dictionary features: Match each multi-world dictionary d to text For token sequences record d_start, d_cont,.. Combine these values with category eg g(h,f)=[places 0 (h)=“places_uniq” and f=organization_start] PittsburghcoachTomlin … Places_uniqPlaces_other Celeb_other Celeb_uniq

Dictionaries in MENE

Borthwick et al: MENE system Features g(h,f) for the loglinear model –External system features: Run someone else’s system s on text For token sequences record sx_start, sx_cont,.. Combine these values with category eg g(h,f)=[proteus 0 (h)=“places_uniq” and f=organization_start]

MENE results (dry run)

MENE learning curves

Largest U.S. Cable Operator Makes Bid for Walt Disney By ANDREW ROSS SORKIN The Comcast Corporation, the largest cable television operator in the United States, made a $54.1 billion unsolicited takeover bid today for The Walt Disney Company, the storied family entertainment colossus. If successful, Comcast's audacious bid would once again reshape the entertainment landscape, creating a new media behemoth that would combine the power of Comcast's powerful distribution channels to some 21 million subscribers in the nation with Disney's vast library of content and production assets. Those include its ABC television network, ESPN and other cable networks, and the Disney and Miramax movie studios. Short names Longer names

LTG system Another MUC-7 competitor Handcoded rules for “easy” cases (amounts, etc) Process of repeated tagging and “matching” for hard cases –Sure-fire (high precision) rules for names where type is clear (“Phillip Morris, Inc – The Walt Disney Company”) –Partial matches to sure-fire rule are filtered with a maxent classifier (candidate filtering) using contextual information, etc –Higher-recall rules, avoiding conflicts with partial-match output “Phillip Morris announced today…. - “Disney’s ….” –Final partial-match & filter step on titles with different learned filter. Exploits discourse/context information

LTG Results

LTG Identifinder (HMMs) MENE+Proteus Manitoba (NB filtered names) NetOwl Commercial RBS

Jansch & Abney paper He was a grammarian, and could doubtless see further into the future than others. -- J.R.R. Tolkien, "Farmer Giles of Ham" echo golf echo x-ray | tr "ladyfinger orthodontics rx-"

Background on JA paper

SCAN: Search & Summarization for Audio Collections (AT&T Labs)

Why IE from personal voic Unified interface for , voic , fax, … requires uniform headers: –Sender, Time, Subject, … –Headers are key for uniform interface Independently, voic access is slow: –useful to have fast access to important parts of message (contact number, caller)

Background on JA – con’t Quick review of Huang, Zweig & Padmanabhan (IBM Yorktown) “Information Extraction from Voic ”: –Goal: find identity and contact number of callers in voic (NER + role classification) –Evaluated three systems on ~= 5000 labeled manually transcribed messages: Baseline system: –200 hand-coded rules based on “trigger phrases” State-of-art Ratnaparki-style MaxEnt tagger: –Lexical unigrams, bigrams, dictionary features for names, numbers, “trigger phrases” + feature selection –Poor results: On manually transcribed data, F1 in 80s for both tasks (good!) On ASR data, F1 about 50% for caller names, 80% for contact numbers even with a very loose performance metric Best learning method barely beat the baseline rule-based system.

What’s interesting in this paper How and when to we use ML? Robust information extraction –Generalizing from manual transcripts (i.e., human-produced written version of voic ) to automatic (ASR) transcripts Place of hand-coding vs learning in information extraction –How to break up task –Where and how to use engineering Candidate Generator Learned filter Candidate phrase Extracted phrase

Voic corpus About 10,000 manually transcribed and annotated voice messages used for evaluation Not quite the usual NER task: we only want the caller’s name

Observation: caller phrases are short and near the beginning of the message.

Caller-phrase extraction Propose start positions i1,…,iN Use a learned decision tree to pick the best i Propose end positions i+j1,i+j2,…,i+jM Use a learned decision tree to pick the best j

Baseline (HZP, Collins log-linear) IE as tagging, similar to Borthwick: Pr(tag i|word i,word i-1,…,word i+1,…,tag i-1,…) estimated via MAXENT model Beam search to find best tag sequence given word sequence (we’ll talk more about this next week) Features of model are words, word pairs, word pair+tag trigrams, …. Hithereit’sBilland… Other Caller_startCaller_contother

Performance

Observation: caller names are really short and near the beginning of the message.

What about ASR transcripts?

Extracting phone numbers Phase 1: hand-coded grammer proposes candidate phone numbers –Not too hard, due to limited vocabulary –Optimize recall (96%) not precision (30%) Phase 2: a learned decision tree filters candidates –Use length, position, a few context features

Results

Their Conclusions