A subtractive medoids selection based fuzzy relational clustering of augmented web user sessions

摘要

Relational Medoid based fuzzy relational clustering (FRC) algorithms perform better than center based FRC. However, in medoid based FRC the selection of medoid is solely random and sometimes lead to inconsistent results. In this paper, a subtractive medoids selection based fuzzy relational clustering (SMS-FRC) method is proposed. In SMS-FRC algorithm inherent geometry and density of pairwise dissimilarity values are preferred over random initial values of medoids. The SMS-FRC is applied to identify clusters of user sessions from server log data, based on their browsing behavior. The concept of augmented sessions is used to derive the page relevance based intuitive augmented dissimilarity matrix. The experiments are performed on a publicly available log data from NASA web server. The generated clusters are evaluated using various fuzzy cluster validity measures, and results are compared with relational fuzzy c-medoids (RFCMdd) clustering algorithm. The results suggest the quality of fuzzy clusters discovered using SMS-FRC clustering is better than that of those obtained with the relational fuzzy c-medoids algorithm.