Abstract

In this paper we introduce a high-precision query classification method to identify the intent of a user query given that it has been seen in the past based on informational, navigational, and transactional categorization. We propose using three vector representations of queries which, using support vector machines, allow past queries to be classified by user's intents. The queries have been represented as vectors using two factors drawn from click-through data: the time users take to review the documents they select and the popularity (quantity of preferences) of the selected documents. Experimental results show that time is the factor that yields higher precision in classification. The experiments shown in this work illustrate that the proposed classifiers can effectively identify the intent of past queries with high-precision. © 2009 Springer.