York University at TREC 2008: Blog Track

York University participated in the TREC 2008 Blog track, by introducing two opinion nding features. By initially focusing solely on the sentiment terms found in a document, using two dierent methods, and not considering the topic relevance, we saw an overall negative impact in the nal evaluations. Post-TREC improvements show the necessity of combining opinion weights with the topic relevance scores. A tunable combination function is used, which revealed some interesting ndings about the opinion nding features noting the strengths and weaknesses. We also introduce the Compass Information Retrieval system, which is based on Okapi, as the driver by which these experiments were conducted.