A Recommendation Model Based on Latent Principal Factors in Web Navigation Data.
Yanzan Zhou, Xin Yu Jin, Bamshad Mobasher · 2004
Discovery of factors that lead to common navigational patterns can help in improving online information presentation as well as in providing personalized content to users. It is, therefore, necessary to develop techniques that can automatically characterize the users ’ underlying navigational objectives and to discover the hidden semantic relationships among users as well as between users and Web objects. Typical approaches to Web usage mining, such as clustering of user sessions, can discover usage patterns directly, but cannot identify the latent factors, intrinsic in users ’ navigational behavior, that lead to such patterns. In this paper, we propose an approach based on a latent variable model, called Iterative Principal Factor Analysis, to discover such hidden factors in Web usage data. The hidden factors are then used to create aggregate models of common user profiles which are, in turn, used to provide dynamic recommendations to users. Our experimental results, performed on real Web usage data, verify that the proposed principal factor approach results in better predictive user models, when compared to more traditional approaches such as clustering and principal component analysis. 1