Knowledgerich Word Sense Disambiguation Rivaling Supervised Systems Presenter
Knowledge-rich Word Sense Disambiguation Rivaling Supervised Systems Presenter : Cheng-Hui Chen Authors : Simone Paolo Ponzetto, Roberto Navigli ACL, 2010 Intelligent Database Systems Lab
Outlines n Motivation n Objectives n Methodology n Experiments n Conclusions n Comments Intelligent Database Systems Lab
Motivation • One of the main obstacles to high-performance WSD is the knowledge acquisition bottleneck. Word. Net WSD algorithm Google ? Knowledge Wiki Other source Intelligent Database Systems Lab
Objectives • To propose to automatically extend Word. Net with large amounts of semantic relations from an encyclopedic resource, name Wikipedia. WSD algorithm Word. Net Resource Wiki Resource Knowledge Intelligent Database Systems Lab
Framework Word. Net resource 1. Synonymy 2. Hypernymy/Hyponymy 3. Sisterhood 4. Gloss Wiki resource 1. Sense labels 2. Links 3. Categories Intelligent Database Systems Lab
Methodology • Wiki – Sense labels – Links – Categories • Word. Net – – Synonymy Hypernymy/Hyponymy Sisterhood Gloss Intelligent Database Systems Lab
Mapping algorithm • SODA (Word. Net) – 1. namely the sodium carbonate • Salt, acetate, sodium, chlorate, benzoate – 2. the drink • Drink, soft, cola, bitter 1. 2. 20 9 • Drink (Wiki) – Soft, drink, cola, sugar Intelligent Database Systems Lab
Experiments • Evaluation of the Mapping Intelligent Database Systems Lab
Experiments • Coarse-grained WSD Ext. Lesk: I am super man Gloss, Sisterhood, Synonymy, Hypernumy/Hyponymy Intelligent Database Systems Lab
Conclusions • This paper has presented a large-scale method for the automatic enrichment of a computational lexicon with encyclopedic relational knowledge. • The method enables simple knowledge-based WSD systems to perform as well as the highest-performing supervised system. Intelligent Database Systems Lab
Comments • Advantages – This paper uses the consistent example. • Applications – Knowledge-based WSD system. Intelligent Database Systems Lab
- Slides: 11