User Interest Profiling Refinement Based On Scientific Paper Keyword Clustering

As the research and development of personalized recommendation techniques and user interest profiling, personalized recommendation has been put into practice in many fields. Among the conventional user profiling algorithms, the ontology-based user profiling methods have gained much attention from researchers for allowing interest inference and many other advantages. However, personalized recommendation based on these traditional ontologies or classifications suffers from their drawbacks like coarse granularity and needing human participation, etc. Thereby personalized recommendation services demands a new user interest profiling method which improves the accuracy and granularity of user interest profiles.To refine the conventional ontology or classification based user interest profiling techniques, this paper focuses on the following three topics mainly:First, we introduce a subject classification refinement method based on keyword concurrence networks. In this method, we use weighted keyword concurrence networks to present the keyword distribution of subjects or research topics. Through using a keyword clustering algorithm in the keyword concurrence networks, we can obtain the refined subject classification system. The refined classification system paves the way for the refinement of user interest profiling.Secondly, a user interest profiling method based on keyword clustering is proposed. We improve the conventional ontology-based interest profiling algorithm by adding a relevance factor measuring the tendency of a user's interest towards the essential content of a certain topic, which makes it easy to differentiate between the subjects of which the user's interest degrees are close. Besides, the user interest profiling algorithm in this paper makes a distinction between explicit interests and implicit interests by looking into the bond between topics or subjects in the classification system. Implicit interest degrees are calculated by taking into account the explicit interest profiles. In application, explicit interest profiling is the foundation of recommendation, and implicit interest profiling serves as a complement. The union of explicit and implicit interest profiles as one entire interest profile improves the accuracy of user interest models.Last but not least, verifying the classification extension and the new user profiling method on scientific paper recommendation system (SPRS), results show that the new methods improve the precision of user interest profiles and the quality of personalized recommendation services.