Comparison of Interestingness Measures for Web Usage Mining: an Empirical Study

A common problem in mining association rules or sequential patterns is that a large number of rules or patterns can be generated from a database, making it impossible for a human analyst to digest the results. Solutions to the problem include, among others, using interestingness measures to identify interesting rules or patterns and pruning rules that are considered redundant. Various interestingness measures have been proposed, but little work has been reported on the effectiveness of the measures on real-world applications. We present an application of Web usage mining to a large collection of Livelink log data. Livelink is a web-based product of Open Text Corporation, which provides automatic management and retrieval of different types of information objects over an intranet, an extranet or the Internet. We report our experience in preprocessing raw log data, mining association rules and sequential patterns from the log data, and identifying interesting rules and patterns by use of interestingness measures and some pruning methods. In particular, we evaluate a number of interestingness measures in terms of their effectiveness in finding interesting association rules and sequential patterns. Our results show that some measures are much more effective than others.