Improving Web Query Classification by Latent Topic Analysis

As Web search engines play an important role in helping people find required information from massive Web data, the Web query classification (WQC) problem becomes an important research issue. Contrast to traditional classification problems, WQC is to classify Web queries into relevant categories of a Web taxonomy. In addition to the typical challenge of processing short and ambiguous queries, WQC faces the practical difficulty of classifying queries into more than one category. In this paper, we propose a semantic-based scheme that uses the Latent Dirichlet Allocation (LDA) model to exploit the latent topic semantics of expanded queries and Web categories for WQC. To evaluate the effectiveness, we have conducted experiments in comparison with the algorithm proposed by Shen et al. in 2005. The results show that our approach has 6.5% and 6.6% improvements in the top-5 classification results for precision and F1, respectively. Keyword: Web Query Classification, Query Expansion, Latent Topic Analysis, Weighted Ensemble