Clustering User Preferences Using W-kmeans

Although commonly only document clustering is suggested by Web mining techniques for recommendation systems, one of the various tasks of personalized recommendation is categorization of Web users. In this paper, a method for clustering navigation patterns of Web users is proposed. We adapt the WordNet-enabled W-kmeans algorithm, an enhancement of standard k-means algorithm which uses the external knowledge from WordNet hypernyms and that has been previously used for document clustering, to user profile clustering by analyzing the users´ historical data. We also investigate the effects this approach has on the recommendation engine by evaluating the overall performance it has in terms of precision -- recall on our online recommendation system.