Feature Matrices: A Model for Efficient and Anonymous Web Usage Mining

Recent growth of startup companies in the area of Web Usage Mining is a strong indication of the effectiveness of this data in understanding user behaviors. However, the approach taken by industry towards Web Usage Mining is off-line and hence intrusive, static, and cannot differentiate between various roles a single user might play. Towards this end, several researchers studied probabilistic and distance-based models to summarize the collected data and maintain only the important features for analysis. The proposed models are either not flexible to trade-off accuracy for performance per application requirements, or not adaptable in real-time due to high complexity of updating the model. In this paper, we propose a new model, the FM model, which is flexible, tunable, adaptable, and can be used for both anonymous and on-line analysis. Also, we introduce a novel similarity measure for accurate comparison among FM models of navigation paths or cluster of paths. We conducted several experiments to evaluate and verify the FM model.