Abstract: Focusing on the problem of low accuracy of similarity measurement and necessarily determining the number of clustering in advance in clustering algorithms of user sessions in existing personalized recommendation services sys- tans, a global alignment method based on dynamic programming algorithm was proposed to measure similarity between user sessions by integrating the information of serialized visiting pages and visiting times. On this basis,automatic spec- tral clustering was done on the user sessions by using improved NJW clustering algorithm. Experimental results show that the algorithm achieves a higher clustering performance than the comparative algorithms by considering the overall characteristics and local information of user sessions. It can also improve the efficiency of Web personalized recommen- lotion services.

Key words: Global alignment, Similarity, User session, Spectral clustering, Automatic clustering