User Profiling: Web Usage Mining

Web usage mining differs from collaborative filtering in the fact that we are not interested in explicitly discovering user profiles but rather usage profiles. When preprocessing a log file we do not concentrate on efficient identification of unique users but rather try to identify separate user sessions. These sessions are then used to form the so called transactions (see [3]). In the following stage, Web usage mining techniques are applied to identify frequent item-sets, sequential patterns, clusters of related pages and association rules (see Sections 3 and 4). Web usage mining can be used to support dynamic structural changes of a Web site in order to suit the active user, and to make recommendations to the active user that help him/her in further navigation through the site he/she is currently visiting. Furthermore, recommendations can be made to the site administrators and designers, regarding structural changes to the site in order to enable more efficient browsing. In the case of implementing Web usage mining system in the form of a proxy server, predictions about which pages are likely to be visited in near future can be made, based on the active users’ behavior. Such pages can be pre-fetched to reduce access times.