Knowledge Discovery from Web Usage Data: Extraction of Sequential Patterns through ART1 Neural Network Based Clustering Algorithm

This paper focuses on extraction of Sequential Patterns (SPs) with very low support from a large preprocessed Web usage data, to discover the behaviors of minority users of a Web site. Due to the sequential nature of the Web user's activity, Sequential Pattern Mining (SPM) is particularly well adapted for the study of Web usage data. Traditional SPM techniques with very low support produce large number of SPs. They are unsuitable for extraction of knowledge about the minority users because of large diversified user's behaviors and difficult to locate. Here, we propose a novel approach called Cluster and Extract Sequential Patterns (CESP) that works based on divisive principle, where initial large Web log data split into smaller clusters(sub-logs) through ART1 neural network based clustering, and then Apriori like SPM technique is applied on each Cluster to extract SPs which reveal the behaviors of minority users. Several experiments were conducted on diversified Web log files, enabled us to discover interesting SPs having very low support (0.06 %). The study reveals that discovery of such SPs by a traditional SPM algorithms were impractical. Keywords: Sequential Patterns, Clustering, Knowledge Discovery, ART Neural Network