Discovering task-oriented usage pattern for web recommendation

Web transaction data usually convey user task-oriented behaviour pattern. Web usage mining technique is able to capture such informative knowledge about user task pattern from usage data. With the discovered usage pattern information, it is possible to recommend Web user more preferred content or customized presentation according to the derived task preference. In this paper, we propose a Web recommendation framework based on discovering task-oriented usage pattern with Probabilistic Latent Semantic Analysis (PLSA) model. The user intended tasks are characterized by the latent factors through probabilistic inference, to represent the user navigational interests. Moreover, the active user’s intuitive task-oriented preference is quantized by the probabilities, by which pages visited in current user session are associated with various tasks as well. Combining the identified task preference of current user with the discovered usage-based Web page categories, we can present user more potentially interested or preferred Web content. The preliminary experiments performed on real world data sets demonstrate the usability and effectiveness of the proposed approach.