Presentation is loading. Please wait.

Presentation is loading. Please wait.

IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved U M I A Searching for knowledge with UIMA IBM Research J.

Similar presentations


Presentation on theme: "IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved U M I A Searching for knowledge with UIMA IBM Research J."— Presentation transcript:

1 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved U M I A Searching for knowledge with UIMA IBM Research J. William Murdock Christopher Welty David Ferrucci Last Update: May 14, 2006

2 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 2 Within-document analysis Person (Entity Annotation) Organization (Entity Annotation) OwnerOf (Relation Annotation) Entity: Person Relation: OwnerOf Entity: Organization Joe Gradgrind, owner of GF,... Person (Entity Annotation) Organization (Entity Annotation) OwnerOf (Relation Annotation) Entity: Person Relation: OwnerOf Entity: Organization Person (Entity Annotation) Joseph Gradgrind, who is the owner of Gradgrind Foods,... doc1.txtdoc2.txt

3 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 3 Cross-document coreference Person (Entity Annotation) Organization (Entity Annotation) OwnerOf (Relation Annotation) Entity: Person Relation: OwnerOf Entity: Organization Joe Gradgrind, owner of GF,... Person (Entity Annotation) Organization (Entity Annotation) OwnerOf (Relation Annotation) Entity: Person Relation: OwnerOf Entity: Organization Person (Entity Annotation) Joseph Gradgrind, who is the owner of Gradgrind Foods,... Relation: OwnerOf Entity: Organization Entity: Person doc1.txtdoc2.txt

4 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 4 EKDB: Extracted Knowledge Database (same information, in relational tables) doc1.txt doc2.txt Referents PersonJoseph Gradgrind OwnerOf OrganizationGradgrind Foods Relation Arguments Documents domainValue rangeValue Person OwnerOf Organization Annotations 016 1821 013 1849 027 4049 2527 Spans Joseph Gradgrind Joe Gradgrind Gradgrind Foods GF Names * Not shown: component ID’s, confidences, etc.

5 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 5 Entity Search Person Joe Gradgrind User query:Subject of interest: Joe Gradgrind Person All persons named “Joe Gradgrind” All entities named “Joe Gradgrind” All persons

6 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 6 Entity Search in EKDB doc1.txt doc2.txt Person OwnerOf Organization domainValue rangeValue Person OwnerOf Organization Joseph Gradgrind Joe Gradgrind Gradgrind Foods GF Person Joe Gradgrind User Query InterfaceEKDB

7 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 7 Browsing entities found by Entity Search Person Joe Gradgrind User query Entities matching the query Names of the entities Joseph Gradgrind Joe Gradgrind H. Joseppi Gradgrind Joe Gradgrind Documents in which the entities occur doc1.txtdoc2.txtdoc88.txt Spans in the documents Joseph Gradgrind, who is the owner of Gradgrind Foods,...... PersonH. Joseppi Gradgrind Facts (relations) involving the entities PersonJoseph GradgrindOrganizationGradgrind FoodsOwnerOf PersonJoseph Gradgrind Browsing facts...

8 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 8 Fact Search Person Joe Gradgrind User query:Subject of interest: Joe Gradgrind Person Some person named “Joe Gradgrind” owns some organization named “Gradgrind Foods” Some entity named “Joe Gradgrind” owns some organization Some person owns something Some relationship from some entity named “Joe Gradgrind” to some entity named “Gradgrind Foods”... Organization Gradgrind Foods Organization OwnerOf Joe GradgrindGradgrind Foods

9 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 9 Fact Search in EKDB doc1.txt doc2.txt Person OwnerOf Organization domainValue rangeValue Person OwnerOf Organization Joseph Gradgrind Joe Gradgrind Gradgrind Foods GF Person Joe Gradgrind User Query InterfaceEKDB OwnerOf Organization Gradgrind Foods

10 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 10 Browsing facts (relations) found by Fact Search User query Facts matching the query Documents in which the facts occur doc1.txtdoc2.txt Spans in the documents Joseph Gradgrind, who is the owner of Gradgrind Foods, Entities involved in the facts PersonJoseph Gradgrind OrganizationGradgrind Foods Browsing entities... Person Joe GradgrindGradgrind Foods PersonJoseph GradgrindOrganizationGradgrind Foods PersonJoseph GradgrindOrganizationGradgrind Foods OwnerOf... ManagerOf

11 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 11 Fact chain search Person Joe Gradgrind City Manchester ??? User query: OrganizationGradgrind Foods CityStockport BasedIn OwnerOf PersonJoe Gradgrind OrganizationGradgrind Foods Subject of interest: Some (complex?) relationship between a person named “Joe Gradgrind” and a city named “Manchester” CityManchester Near CityStockport NationEngland NationEngland SubPlace CitizenOf PersonJoe Gradgrind CityManchester

12 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 12 Fact pattern search Person Organization User query: A person that that resides in Leeds and owns an organization in Stockport City Leeds Subject of interest: ResidesIn OwnerOf City Stockport BasedIn

13 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 13 Status  Entity Search & Fact Search implemented in SAW 1 –But limited interaction between the two –Thus misses some of the recursive nature of browsing entities and facts (entities participate in facts, that contain entities, etc.)  Prototype of Fact Chain Search implemented in a SAW 1 variant –No metrics for “interestingness” of chains yet  Fact Search implemented in SAW 2 –More capabilities on the way  Fact Pattern Search: Future work

14 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 14 Backup Slides

15 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 15 Relations  Facts  We extract “relations” in the ACE sense: concrete instances of relationships. –We usually use the term “relations” when talking about extraction.  End users have found that term confusing, so in user interfaces we prefer the term “facts.” –We usually use the term “facts” when talking about search. OrganizationGradgrind Foods OwnerOf PersonJoe Gradgrind

16 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 16 Who is Joseph Gradgrind? Thomas Gradgrind, sir. A man of realities. A man of fact and calculations. A man who proceeds upon the principle that two and two are four, and nothing over, and who is not to be talked into allowing for anything over.... You might hope to get some other nonsensical belief into the head of George Gradgrind, or Augustus Gradgrind, or John Gradgrind, or Joseph Gradgrind (all supposititious, non-existent persons), but into the head of Thomas Gradgrind -- no, sir! - Charles Dickens, Hard Times, Chapter 2 A fictional, suppositious, non-existent British person

17 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 17 Motivation: Why Transform Knowledge?  Different systems have different ontologies and/or different representational schemes  Sometimes those differences are arbitrary  Other times they are specifically motivated by differences in the purposes of the systems  In either case, interoperation requires that knowledge be transformed

18 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 18 Knowledge Integration in UIMA: Overview The transformation of knowledge from one form to another requires the explicit mapping across ontologies. Relation (ManagerOf) Entity (Person): Fred Center Entity (Organization): Center Micros Executive: Fred Center SocialAggregate: Center Micros hasManager Organization(?x)  SocialAggregate(?x)Person(?x) ^ ManagerOf(?x, ?y)  Executive(?x) KITE Mapping Plugins ManagerOf(?x, ?y)  hasManager(?y, ?x) Target Ontology Source Ontology

19 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 19 KITE-based applications Source Plugin Ontology Language Plugin Mapper Plugin(s) Target Plugin Ontology Language Plugin Source Data Target Data Provenance Plugin Source Repository Target Repository Provenance Repository Source Ontology Target Ontology

20 IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved 20 Building KITE applications  Framework provides: –API’s for: Mapper plugins Source plugins Target plugins Provenance plugins Language plugins –Classes for Data –Top-level control from source  mapper  target –Some broadly applicable plugins (of each of the types)  Application developer provides: –Configuration for some of KITE’s broadly applicable plugins –New, application specific plugins (if needed) Source Plugin Ontology Language Plugin Mapper Plugin(s) Provenance Plugin Target Plugin Ontology Language Plugin


Download ppt "IBM Research | Semantic Analysis and Integration © 2006 IBM Corporation – All Rights Reserved U M I A Searching for knowledge with UIMA IBM Research J."

Similar presentations


Ads by Google