|
|
MODELING AND EXPLAINING SIMILARITY
Patrick Pantel USC Information Sciences Institute Fri, October 26, 3:30pm, MJH Rm 126 Similarity modeling is a key task in computational lexical semantics for finding word senses, concepts, paraphrases, topics, and distributional synonyms, just to name a few. Missing, however, are ways to automatically provide explanations of why a system finds two elements similar. In this talk, we will survey the state of the art in large-scale semantic similarity modeling, focusing on methods for mapping problem statements to feature representations, information theoretic feature weighting, comparison measures, and clustering algorithms. Then, we will present our recent work on automatically finding explanations for why elements are similar and discuss plans for using these to build an interactive similarity platform. 
|