Semantic Outlier Analysis for Sessionizing Web Logs

초록

As the web usage patterns from clients are getting more complex, simple sessionizations based on time and navigation-oriented heuristics have been restricted to exploit various kinds of rule discovering methods. In this paper, we present semantic session reconstruction based on semantic outliers from web log data. Above all, web directory service such as Yahoo is applied to enrich semantics to web logs, as categorizing them to all possible hierarchical paths. In order to detect the candidate set of session identifiers, semantic factors like semantic mean, deviation, and distance matrix are established. Eventually, each semantic session is obtained based on nested repetition of top-down partitioning and evaluation process. For experiment, we applied this ontology-oriented heuristics to sessionize the access log files for one week from IRCache. Compared with time-oriented heuristics, more than 48% of sessions were additionally detected by semantic outlier analysis. It means that we can conceptually track the behavior of users tending to easily change their intentions and interests, or simultaneously try to search various kinds of information on the web.