UniNE at TREC 2008: Fact and Opinion Retrieval in the Blogsphere

This paper describes our participation in the Blog track at the TREC 2008 evaluation campaign. The Blog track goes beyond simple document retrieval, its main goal is to identify opinionated blog posts and assign a polarity measure (positive, negative or mixed) to these information items. Available topics cover various target entities, such as people, location or product for example. This year’s Blog task may be subdivided into three parts: First, retrieve relevant information (facts & opinionated documents), second extract only opinionated documents (either positive, negative or mixed) and third classify opinionated documents as having a positive or negative polarity. For the first part of our participation we evaluate different indexing strategies as well as various retrieval models such as Okapi (BM25) and two models derived from the Divergence from Randomness (DFR) paradigm. For the opinion and polarity detection part, we use two different approaches, an additive and a logistic-based model using characteristic terms to discriminate between various opinion classes.