NSF-ITR/IM PROJECT: 2004 Abstracts From Bits to Information: Statistical Learning Technologies for Digital Information Management Search

Project Title: Term Informativeness PI: T. Jaakkola Participants: Jason Rennie and Tommi Jaakkola (MIT CSAIL) Abstract: Informal communication (e-mail, bulletin boards) poses a difficult learning environment because traditional grammatical and lexical information are noisy. For named entity extraction how topic-centric, or “informative,” a word is can provide valuable additional information. We introduce a new informativeness score based on mixture models for the task of extracting restaurant names from bulletin board posts. By combining the mixture score with IDF, we are able to achieve significant gains on a restaurant extraction task. We also motivate and discuss a Bayesian version of the score which would better capture the variability in term occurrence rates.