Relational clustering based on a new robust estimator withapplication to Web mining

摘要

Mining typical user profiles and URL associations from the vastamount of access logs is an important component of Web personalization.In this paper, we define the notion of a "“user session” asbeing a temporally compact sequence of Web accesses by a user. We alsodefine a dissimilarity measure between two Web sessions that capturesthe organization of a Web site. To cluster the user sessions based onthe pairwise dissimilarities, we introduce the relational fuzzyc-maximal density estimator (RFC-MDE) algorithm. RFC-MDE is robust andcan deal with outliers that are typical in this application. We showreal examples of the use of RFC-MDE for extraction of user profiles fromlog data, and and compare its performance to the standard non-Euclideanfuzzy c-means