Clustering Web Access Patterns Based on Hybrid Approach

The interest of web users can be revealed by the visited web pages and time duration on these web pages during their surfing. In this paper, each web access pattern from web logs is transformed into a fuzzy vector with predetermined dimension, each component being a fuzzy linguistic variable or 0 representing the visited web page and the time duration on this web page. Fuzzy simulation is used to compute the distance between any two fuzzy vectors. Considering the clustering time and efficiency, we propose an evolutionary two-layer clustering algorithm. At the first layer, the learning vector quantization (LVQ) approach is exploited to group the patterns from web logs into a number of clusters. At the second layer, the weighted fuzzy c-means approach is developed to deal with the results of the first layer. In addition, PSO algorithm is adopted to optimize the clustering results. The effectiveness and feasibility of the approach are demonstrated by the algorithm analysis and our experimental results.