joint posterior revision of nlp annotations via
play

Joint Posterior Revision of NLP Annotations via Ontological - PowerPoint PPT Presentation

#4479 Joint Posterior Revision of NLP Annotations via Ontological Knowledge Marco Rospocher Francesco Corcoglioniti Context: Knowledge Extraction Kia has hired Peter Schreyer as chief design officer. Joint Posterior Revision of NLP


  1. #4479 Joint Posterior Revision of NLP Annotations via Ontological Knowledge Marco Rospocher Francesco Corcoglioniti

  2. Context: Knowledge Extraction Kia has hired Peter Schreyer as chief design officer. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  3. Context: Knowledge Extraction Kia has hired Peter Schreyer as chief design officer. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  4. Context: Knowledge Extraction Organization Kia has hired Peter Schreyer as chief design officer. NLP Tasks: - Named Entity Recognition and Classification (NERC) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  5. Context: Knowledge Extraction Organization dbpedia:Kia_Motors Kia has hired Peter Schreyer as chief design officer. NLP Tasks: - Named Entity Recognition and Classification (NERC) - Entity Linking (EL) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  6. Context: Knowledge Extraction Organization dbpedia:Kia_Motors Kia has hired Peter Schreyer as chief design officer. framenet:employer NLP Tasks: - Named Entity Recognition and Classification (NERC) - Entity Linking (EL) - Semantic Role Labeling (SRL) … Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  7. Motivating Examples Mr. Washington was runner-up at Wimbledon in 1996. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  8. Motivating Examples Mr. Washington was runner-up at Wimbledon in 1996. http://nlp.stanford.edu:8080/corenlp Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  9. Motivating Examples Mr. Washington was runner-up at Wimbledon in 1996. http://nlp.stanford.edu:8080/corenlp http://demo.dbpedia-spotlight.org http://dbpedia.org/resource/ Washington_(state) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  10. Motivating Examples Mr. Washington was runner-up at Wimbledon in 1996. http://nlp.stanford.edu:8080/corenlp http://demo.dbpedia-spotlight.org http://dbpedia.org/resource/ Washington_(state) The GW Bridge is a double-decked suspension bridge over the Hudson. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  11. Motivating Examples Mr. Washington was runner-up at Wimbledon in 1996. http://nlp.stanford.edu:8080/corenlp http://demo.dbpedia-spotlight.org http://dbpedia.org/resource/ Washington_(state) The GW Bridge is a double-decked suspension bridge over the Hudson. http://demo.dbpedia-spotlight.org http://dbpedia.org/resource/ George_Washington_Bridge Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  12. Motivating Examples Mr. Washington was runner-up at Wimbledon in 1996. http://nlp.stanford.edu:8080/corenlp http://demo.dbpedia-spotlight.org http://dbpedia.org/resource/ Washington_(state) The GW Bridge is a double-decked suspension bridge over the Hudson. http://demo.dbpedia-spotlight.org http://nlp.stanford.edu:8080/corenlp http://dbpedia.org/resource/ George_Washington_Bridge Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  13. Abstracting … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  14. Abstracting … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  15. Abstracting Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  16. Abstracting a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  17. Abstracting a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  18. Abstracting a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  19. RESEARCH PROBLEM How can we assess and improve the coherence of the various NLP annotations on an entity mention?

  20. In a nutshell ontological background knowledge a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  21. In a nutshell ontological background knowledge a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  22. In a nutshell ontological background knowledge a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  23. In a nutshell ontological background knowledge a 1,1 a 2,1 a n,1 a 1,2 a 2,2 a n,2 … … … a 1,k a 2,i a n,j Task 1 Task n Task 2 … token 1 token 2 token 3 token 4 token 5 token 6 …. Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  24. Contributions 1. JPARK: a probabilistic model capable to estimate a posteriori the overall confidence of NLP annotations 2. A concrete instantiation of the model for NERC and EL (using YAGO as ontological knowledge) 3. Application of the NERC and EL model to revise the annotations of Stanford NER and DBpedia Spotlight Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  25. The Model P ( a | m , B , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  26. The Model entity mention NLP Background Knowledge ( a i , … , a n ) NLP Annotations “The” Ontological Knowledge P ( a | m , B , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  27. The Model entity mention NLP Background Knowledge ( a i , … , a n ) NLP Annotations “The” Ontological Knowledge P ( a | m , B , K ) set of classes from K P ( a ,C | m , B , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  28. The Model entity mention NLP Background Knowledge ( a i , … , a n ) NLP Annotations “The” Ontological Knowledge P ( a | m , B , K ) set of classes from K P ( a ,C | m , B , K ) P ( a i | m , B ) P ( C | a i , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  29. The Model entity mention NLP Background Knowledge ( a i , … , a n ) NLP Annotations “The” Ontological Knowledge P ( a | m , B , K ) set of classes from K P ( a ,C | m , B , K ) confidence score P ( a i | m , B ) P ( C | a i , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  30. The Model entity mention NLP Background Knowledge ( a i , … , a n ) NLP Annotations “The” Ontological Knowledge P ( a | m , B , K ) set of classes from K P ( a ,C | m , B , K ) confidence score learned from data P ( a i | m , B ) P ( C | a i , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  31. The Model entity mention NLP Background Knowledge ( a i , … , a n ) NLP Annotations “The” Ontological Knowledge = arg max a P ( a | m , B , K ) set of classes from K P ( a ,C | m , B , K ) confidence score learned from data P ( a i | m , B ) P ( C | a i , K ) Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  32. NERC and EL Model

  33. Ingredients • Ontological Knowledge • P ( C | a NERC , K ) Estimating Estimating P ( C | a EL , K ) • Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

  34. Ingredients • Ontological Knowledge • P ( C | a NERC , K ) Estimating Estimating P ( C | a EL , K ) • Joint Posterior Revision of NLP Annotations via Ontological Knowledge - M. Rospocher & F. Corcoglioniti

Recommend


More recommend