Clustering of Web Users Based on Competitive Agglomeration

An important component of web personalization is to mine typical user profiles from the vast amount of historical data stored in access logs. A new clustering algorithm based on user transactions was proposed to provide personalized recommendation service for the websites. As an improvement on K-means algorithm, we got best cluster number and initial clustering centers automatically by competitive agglomeration, we established access matrix based on the access sequence, browsing time, click frequency. A new distance method that captures the structure of a web site is defined to measure the similarity between two users. We exploited the definition of clustering centers in k-path algorithm and enhanced CAKPS algorithm to cluster the access users. Experiments are performed to compare the CAKPS algorithm with two other algorithms, and the results show that the enhanced algorithm convergences more rapidly and the difference of user sets are higher than the normal algorithm.