Log Mining to Improve the Performance of Site Search

Despite of the popularity of global search engines, people still suffer from low accuracy of site search. The primary reason lies in the difference of link structures and data scale between global Web and website, which leads to failures of traditional re-ranking methods such as HITS, PageRank and DirectHit. This paper proposes a novel re-ranking method based on user logs within websites. With the help of website taxonomy, we mine for generalized association rules and abstract access patterns of different levels. Mining results are subsequently used to re-rank the retrieved pages. One of the advantages of our mining algorithm is that it resolves the diversity problem of user’s access behavior and discovers general patterns. Experiment shows that the proposed method outperforms keyword-based method by 15% and DirectHit by 13% respectively.