Exploring Traits of Adjectives to Predict Polarity Opinion in Blogs and Semantic Filters in Genomics

This paper presents the results of our team in the Genomics and Blog tracks in TREC 2007. We used the language model implementation provided by Indri for both tracks. For the BLOG track we explored the use of adjectives with in a post as a way to predict opinion polarity. Our work in the Genomics track explores two approaches to generate queries from the original topics. The first approach performs automatic term expansion using UMLS to generate a structured query that can be submitted using Indri’s query language. The second approach uses a query expansion and re-ranking method based on identification of semantic relatives. This approach tries to capture the semantic of the potential answer, key terms in the topic and detection of gene/protein terms mentioned in the topic.