Global and Local Models for Multi-document Summarization Pradipto Das and Rohini Srihari SUNY Buffalo TAC 2011, Gaithersburg, MD
Global and Local Models Local Models Accidents and Natural Disasters Endangered Resources Attacks Global (Criminal/Terroris Model Investigations t) and Trials (Criminal/Legal/ Other) Multi-document Health and summaries Safety
An Example of a Global Model Topic Translation Topic Translation Topic Translation सुनामी , पवमान , चीन , Tsunami, flight, China, Topics over words भूक ं प , चाइऱ , एयर , फ़ॎांस , ओऱंपपक , earthquake, Air, France, Olympic, पपचचऱेमू , जहाज़ , बीजजंग , Chile, Brazil, Beijing, गये , ब्ऱाज़ीऱ , गोर , समारोह , Pichilemu, A, 447, Gore, चेतावनी , ए , ४४७ , गायब , सॎवरॎण , gone, disappear, function, खबर , महासागर , सॎटेडियम , warning , ocean stadium, शहर फ़ॎांसीसी खेऱोः news, city France games Topics over controlled vocabulary Topic Translation Topic Translation Topic Translation सुनामी , ब्ऱाज़ीऱ , ए , चीन , ओऱंपपक Tsunami, Brazil, A, China, भूक ं प , गायब , खोज , चीन : xx->xx, earthquake, disappeared, Olympic, उड़ान , बीजजंग , भूक ं प : xx->xx, earthquake:x search, flight, China:xx->xx, शहर , पवमान : xx- ओऱंपपक : xx- x->xx, city, aircraft:xx- Olympic:xx- सॎथानीय , >xx, गोर : xx- local, UTC, >xx, >xx, ocean, >xx, Gore:xx- यू०टी०सी० , महासागर , >xx, गोर , Mayor, ship:xx->xx, >xx, Gore, मेयर , जहाज़ : xx->xx, सॎवरॎण , Tsunami:xx- air:xx->xx, gold, सुनामी :xx->xx एयर : xx->xx, बीजजंग : xx- >xx air, space Beijing:xx- हवाई , ऺेत्ऱ >xx, नेशनऱ >xx, National
Bi-Perspective Document Structure Words in Para 1 Manually Words edited Wiki category tags in – words that Para 2 summarize/ categorize the document Wikipedia
Understanding the Two Perspectives Imagine browsing over reports in a topic cluster It is believed US investigators have asked for, but have been so far refused access to, evidence accumulated by German prosecutors probing allegations that former GM director, Mr. Lopez, stole industrial secrets from the US group and took them with him when he joined VW last year. This investigation was launched by US President Bill Clinton and is in principle a far more simple or at least more single-minded pursuit than that of Ms. Holland. Dorothea Holland, until four months ago was the only prosecuting lawyer on the German case. News Article
Understanding the Two Perspectives What words can we remember after a first browse? It is believed US investigators have asked for, but have been so far refused access to, evidence accumulated by German prosecutors probing allegations that former GM director , Mr. German, US, Lopez , stole industrial secrets from the US group investigations, and took them with him when he joined VW last year. GM, Dorothea This investigation was launched by US Holland, Lopez, President Bill Clinton and is in principle a far more simple prosecute or at least more single-minded pursuit than that of Ms. Holland . The “document level” Dorothea Holland , until four months ago perspective was the only prosecuting lawyer on the German case. News Article
Understanding the Two Perspectives What helped us generate the Document Level perspective? The “word level” perspective It is believed US investigators have asked for, but have been so far refused access to, evidence accumulated by German German, US, prosecutors probing allegations that former Named Entities GM director, Mr. Lopez, stole industrial investigations, LOCATION secrets from the US group and took them GM, Dorothea MISC with him when he joined VW last year. Holland, Lopez, ORGANIZATION This investigation was launched by US President Bill Clinton and is in principle a far PERSON prosecute more simple or at least more single-minded pursuit than that of Ms. Holland. Important Verbs The “document level” and Dependents Dorothea Holland, until four months ago perspective WHAT was the only prosecuting lawyer on the HAPPENED? German case. News Article
What if we turn the document off? Summarization power of the perspectives It is believed US investigators have asked for, but have been so far refused access to, evidence accumulated by German German, US, prosecutors probing allegations that former GM director, Mr. Lopez, stole industrial investigations, secrets from the US group and took them GM, Dorothea with him when he joined VW last year. Holland, Lopez, This investigation was launched by US President Bill Clinton and is in principle a far prosecute more simple or at least more single-minded pursuit than that of Ms. Holland Dorothea Holland, until four months ago was the only prosecuting lawyer on the German case.
Assumptions of the Global Models • Documents are at least tagged from two different perspectives – either implicit or explicit and one perspective affects the other – Simplest example of implicit WL tagging – binned positions indicating sections – Simplest example of implicit DL tagging – tag cloud Begin (0) It is believed US investigators have asked for, but have been so far refused access to, evidence accumulated by German prosecutors probing allegations that former GM director, Mr. Lopez, stole industrial secrets from the US group and took them with him when he joined VW last year. This investigation was launched by US President Bill Clinton and is in principle le (1) Midd a far more simple or at least more single-minded pursuit than that of Ms. Holland. Dorothea Holland, until four months ago was the only prosecuting lawyer on End (2) tagcrowd.com the German case.
Document Level Perspectives • Guided Summarization Track Centers of Attentions (with regard to Menu_foods:ne->ne, pet:nn->nn, grammatical or semantic roles) unit:nn->nn, Henderson:ne->ne, wheat:nn->nn, food:subj->nn etc. Top 20 (tf-idf) docset words + Menu_Foods, pet, associate, plant, sell, Top 5 most frequent non-stopwords in source, FDA, Henderson, agency, shelf, the documents test, unit, Canadian, dog, food etc. • Multilingual Track उतॎतर (North) : xx->xx, घायऱ (injured) : xx- Centers of Attentions (without regard to >xx, जांच (investigation):xx->xx, ऱंदन ( grammatical or semantic roles) London) : xx->xx, पुलऱस (police) : xx->xx etc. जांच (investigation), घरोः (houses), Top 20 (tf-idf) docset words + तऱाशी (search), पुलऱस (police), Top 5 most frequent non-stopwords in ं गॎस (King’s), सॎटेशन (station), कक the documents क्ऱॉस (Cross), हमऱे (attack)etc. Multilingual stopwords found by Google translate
Word Level Perspectives • Guided Summarization Track – Named Entity classes (Person, Organization, Location, Misc, Date/time/money/number/ordinal/percent) – Subjective class e.g. “Of the 10 cats and dogs whose deaths have been linked to pet food that was recalled prep_of - X over the weekend, seven died last month in a taste test conducted by…” Nsubj - √ • Multilingual Track – {0, 1, 2, 3, 4}: Words annotated by positional bins – document segregated into 5 “sections”
Global(Background) Models • METag 2 LDA: A topic generating all DL tags in a The idea was to document doesn’t necessarily mean that the same assign weights topic generates all words in the document to words in • CorrMETag 2 LDA: A topic generating * all* DL tags in a sentences from document does mean that the same topic generates a generative all words in the document standpoint METag 2 LDA CorrME- Topic concentration parameter Tag 2 LDA Document specific topic proportions Indicator variables Document content words Document Level (DL) tags Word Level (WL) tags Topic Parameters Tag Parameters
Global(Background) Models METag 2 LDA CorrMETag 2 LDA पवमान पवमान :xx xx U.S.:nn obj U.S. PER, ORG etc.. 0/1/2 etc..
Recommend
More recommend
