University of Waterloo at TREC 2008 Blog Track
WATERLOO UNIV (ONTARIO) DEPT OF MANAGEMENT SCIENCES
Pagination or Media Count:
The paper reports the University of Waterloo participation in the opinion and polarity tasks of the Blog track. The proposed method uses a lexicon built from several linguistic resources. The opinion discriminating ability of each subjective lexical unit was estimated using the Kullback-Leibler divergence. The KLD scores of subjective words occurring within fixed-size windows around instances of query terms were used in calculating document scores. The described system also used a method of identifying phrases in topic titles by matching them to Wikipedia titles. The results show that both KLD-based scores of subjective lexical units and Wikipedia-matched phrases are useful techniques that help improve opinion retrieval performance.
- Information Science