Query clustering using user logs

Reviewer: Dimitrios Katsaros

The problem of identifying frequently asked questions (FAQs) in information retrieval systems has been previously addressed by clustering the queries into groups according to their similarity. This clustering has been based either on the contents of the queries (clustering based on query content), or solely on the documents that users select as the answer to their query (clustering based on cross-references). This paper proposes a new query clustering method that is a hybrid of the aforementioned approaches. The major contribution of this work is its demonstration of the fact that exploiting only the contents of the query, or only users’ judgment, cannot give correct results in all cases. The paper shows that keyword matching fails to cluster queries effectively, because of the ambiguity of words; and clustering based on cross-references also fails, because it creates query clusters with extremely broad topics. The major weakness of the paper is that it does not demonstrate whether these findings also apply in the case of search engines for the Web, which is both the ultimate challenge and the stated objective of the paper. The results presented in the paper are drawn from a commercial online encyclopedia, and since the encyclopedia’s contents are relatively coherent compared to the Web, it is natural to discover query clusters in this limited venue. The paper is well written, and requires no special background in order to understand the ideas presented, although basic knowledge of information retrieval terminology would be very helpful. Overall, the paper is an interesting case study for query clustering, but it cannot answer the question of whether or not the proposed methods apply in an enormous, dynamic information retrieval system like the Web. Online Computing Reviews Service