An Unsupervised Method for Web User Interest Analysis

In this paper, we propose an unsupervised method to determine dominating user interests from browsing behavior data. This method identifies group of users with similar interests by partitioning a similarity graph, in which nodes are users and edges represent the similarities in browsing behavior between the users. Two users are considered similar if they have viewed the same content. We consider and compare several types of similarity measures, which are calculated at different granularity levels of content: at web page level, page category level, and topic level as identified by topic modeling techniques. A hierarchical graph clustering algorithm is used to perform partitioning process to capture the natural hierarchy of user clusters and producing intuitive features for understanding the content linked to each cluster. For evaluation, we perform an experiment on a browsing behavior dataset from a university's Web portal. Our system can discover interesting information like web user interests, who is interested in which content.