Fusing Blog Opinion Retrieval Results for Better Effectiveness

In recent years, blogs have been very popular on the Web as a grassroots publishing platform. Some research has been conducted on them and blog opinion retrieval is one of the key issues. In this paper, we investigate if data fusion can be useful for improvement of effectiveness of blog opinion retrieval. Extensive experimentation with the runs submitted to the blog opinion retrieval task in TREC 2008 is carried out and a few data fusion methods including Comb Sum, Comb MNZ, Borda count, and the linear combination method are investigated. We observe that generally speaking, all data fusion methods involved are very competitive compared with the best component retrieval system. Especially, the linear combination method with proper training is superior to other data fusion methods and it is able to beat the best component retrieval system by a clear margin. This study demonstrates that data fusion can be an effective technique for blog opinion retrieval if proper fusion methods are applied.