Collaborative filtering recommendation algorithm based on user preference derived from item domain features

Personalized recommendation is an effective method for fighting “information overload”. However, its performance is often limited by several factors, such as sparsity and cold-start. Some researchers utilize user-created tags of social tagging system to depict user preferences for personalized recommendation, but it is difficult to identify users with similar interests due to the differences between users’ descriptive habits and the diversity of language expression. In order to find a better way to depict user preferences to make it more suitable for personalized recommendation, we introduce a framework that utilizes item domain features to construct user preference models and combines these models with collaborative filtering (CF). The framework not only integrates domain characteristics into a personalized recommendation, but also aids to detecting the implicit relationships among users, which are missed by the conventional CF method. The experimental results show our method achieves the better result, and prove the user preference model is more effective for recommendation.