Evaluation of User Model Using Partial Order Relation

The traditional personalized recommendation system measures the quality of recommendation performance by the indicators of precision and recall. User model based on user browsing behavior plays an increasingly important role in personalized recommendation systems. The evaluation of user model is significant for personalized recommendation systems, which mainly focuses on the contents of subjects. Taking into account user's priority for the subjects, we urgently need to make an evaluation for the performance of user model. In this paper, an evaluation framework on performance of priority in user model is introduced, and the similarity calculation method of interest structure is designed. According to this, the evaluation method of stability of update mechanism for user model over an interval is proposed. Ten users verify the similarity of interest priority between their user models and actual interests by priority precision (p_pre) and priority recall (p_rec), and test the degrees of stability of user models for two months in the experiment when users' interests changed continually. The results show that the proposed method is more successful than traditional method in the measurement of interest priority.