A Tag-based Recommendation Algorithm Integrating Short-term and Long-term Interests of Users

Abstract

In the collaborative tagging system, tags of users contain rich information on personalized preference, and time stamps of users show their interest changes. Users’ interest has the varied timeliness, but the existing recommendation algorithm fusing time information only emphasizes the short-term interest of users but fails to dig into the long-term stable interest of users and thus presents the low recommendation precision. In this paper, we propose a tag-based recommendation algorithm integrating short-term and long-term interests of users. The algorithm first builds the short-term and long-term interest characteristics of users based on tags according to the use frequency, time decay, life cycle and volatility; then it maps the characteristics to the resources tagged by the user to form a user-resource pseudo-scoring matrix. Finally, it calculates the set of the nearest neighbors of user with the matrix to give recommendations. The verification on the delicious data set shows that the algorithm improves recommendation precision and diversity compared with other classical algorithms.