Difference between revisions of "Hovy et al., EMNLP 2009"
PastStudents (talk | contribs) (Created page with '== Citation == Eduard Hovy, Zornitsa Kozareva, and Ellen Riloff. 2009. Toward completeness in concept extraction and classification. In Proceedings of the 2009 Conference on Emp…') |
PastStudents (talk | contribs) m |
||
| Line 22: | Line 22: | ||
=== Evaluation === | === Evaluation === | ||
| − | The evaluation was done using the 1000 web snippets returned from Google | + | The evaluation was done using the 1000 web snippets returned from Google on two concepts: lions for Animals and Madonna for People. The results were evaluated by human. |
| − | Compared with WordNet, the proposed automatic method performed comparable in precision on Animal and better on People. In both cases, the coverage | + | Compared with WordNet, the proposed automatic method performed comparable in precision on Animal and better on People. In both cases, the coverage was much better with the proposed method. |
== Related papers == | == Related papers == | ||
A similar idea of using the hyponym patterns was used in KnowItAll system (Etzioni et al., JAI 2005). | A similar idea of using the hyponym patterns was used in KnowItAll system (Etzioni et al., JAI 2005). | ||
Revision as of 14:57, 30 November 2010
Citation
Eduard Hovy, Zornitsa Kozareva, and Ellen Riloff. 2009. Toward completeness in concept extraction and classification. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 2 - Volume 2 (EMNLP '09), Vol. 2. Association for Computational Linguistics, Morristown, NJ, USA, 948-957.
Online version
Summary
This paper proposed a method to automatically extract terms that are related to the given concepts and their taxonomic organization (is-a links). Essentially, the authors try to build the taxonomy below the given concept. The method started with a simple surface pattern with a few open positions and fill these positions differently to extract terms and their taxonomic links.
- The first step of the method generate more seed terms using the Doubly-Anchored Pattern (DAP)(Kozareva et al., ACL-HLT 2008).
- DAP: [seed1] such as [seed2] and [X]
- These terms were then ranked based on how frequently they can be used to find more seed terms.
- Then the Doubly-Anchored Pattern was used in the 'backward' direction to find the higher-level seed.
- Backward DAP: [X] such as [seed1] and [seed2]
- The higher-level terms are filter based on if they can be used to find more lower level terms and if they are below the given root concept.
The above two steps are executed alternatively and the text were web pages returned from Google.
Evaluation
The evaluation was done using the 1000 web snippets returned from Google on two concepts: lions for Animals and Madonna for People. The results were evaluated by human.
Compared with WordNet, the proposed automatic method performed comparable in precision on Animal and better on People. In both cases, the coverage was much better with the proposed method.
Related papers
A similar idea of using the hyponym patterns was used in KnowItAll system (Etzioni et al., JAI 2005).
