Evaluating query classification algorithms.

Abstract

Query clustering and query classification aim to capture the intended meaning of queries in order to improve the efficiency of search engines. This research explored the relationship between the query clustering and the query classification approaches and proposed new query classification approaches named Single Learning and Iterative Learning. The intuition of the Single Learning approach is that given a set of queries categorized under a subject, the subject then can be defined as a cluster of queries. The query cluster gives the semantic description for the subject. If a query Q is similar to a cluster of queries categorized under a subject S, then the query Q can be categorized under the subject S. The Single Learning approach makes use of the Web as a knowledge base to build up the term-weight vectors representing the queries. The subject can then be represented by the query cluster-centroid vector. The similarity SIM (S,Q) between the query cluster S and the query Q is computed and compared to a threshold Teta. If SIM(S,Q) >= Teta then the query Q is categorized under the subject S with the degree of Teta.