Behavioral Patterns Mining for Online Time Personalization

Behavioral patterns represent repeating sequences or sets of actions, which website users often perform together. Such patterns can be used to identify user preferences, recommend interesting content to him, etc. For dynamic sites with fast changing content (e.g., news, social networks) we need to recognize such patterns in an online time. In this paper, we introduce a novel method for recognizing behavioral patterns in an online time over a data stream. Main contribution is a combination of global patterns with patterns specific for groups of similar users. We evaluated the method using a personalized recommendation task over datasets from news and e-learning domains and show that the combination of common global and specific group patterns reaches higher recommendation precision than its components used individually.