DEWS2008 B3-5 Finding Latent Neighbors for Query Recommendation: a User-Controllable Scheme

Abstract Ambiguous queries and the large amount of unindexed information by search engines give rise to the problem of finding latent neighbors for query recommendation. In this paper, we give our definition for latent neighbors of a query, and then propose a novel two-phase algorithm to solve this problem, which takes the con-nectivity of the query-URL bipartite graph as input. In addition, due to the subjectivity of similarity measures, a user-controllable scheme is presented to bridge the gap between the determinacy of similarity values and the inde-terminacy of users ’ information needs. The experiment results from two mixtures of data collections demonstrate the usefulness of our algorithm and the feasibility of our scheme. Key words query recommendation, clustering, rank mechanism 1