作者: Edward DeGuzman , Victor Lavrenko , James Allan , Veera Pollard , Stephen Thomas
DOI:
关键词:
摘要: We extend relevance modeling to the link detection task of Topic Detection and Tracking (TDT) show that it substantially improves performance. Relevance modeling, a statistical language technique related query expansion, is used enhance topic model estimate associated with news story, boosting probability words are story even when they do not appear in story. To apply TDT, had be extended work stories rather than short queries, similarity comparison changed modified form Kullback-Leibler. demonstrate models result very substantial improvements over baseline. also how use makes possible choose single parameter for within- cross-mode comparisons stories.