University of Lugano at TREC 2008 Blog Track

摘要

We report on the University of Lugano's participation in the Blog track of TREC 2008. In particular we describe our system for performing opinion retrieval and blog distillation. The 2008 Blog track continued on from the successful 2007 Blog track, including the same opinion retrieval and blog distillation activities. This year was our first participation in TREC and we participated in both opinion retrieval and blog distillation tasks. We aimed to test the effectiveness of learning methods in each of these tasks. In the topic retrieval phase (baseline) of the opinion retrieval task, we used a rank learning method to combine additional information including the content of incoming hyperlinks and tag data from social bookmarking websites with our basic retrieval method which was the Divergence from Randomness version of BM25 (DFR BM25). The results shows 20% improvement in the Mean Average Precision (MAP) of the proposed method in comparison with DFR BM25. We then examined the effectiveness of learning methods in assigning opinion scores to documents. We compared a Support Vector Machine (SVM) based learning system with a simpler system that used the average opinionatedness of each word in the document. Although the results were not satisfactory in our TREC submission and we didn't improve the baseline, repeating the experiments showed the improvement over the baseline by using the learning methods in document opinion scoring.