Effective recommendation is indispensable to customized or personalized services. The ease of collecting, integrating and analyzing vast amounts of data about customers and their purchase intentions/behaviors concerning different products or services has greatly fostered the interest in automated recommendations appropriate for individual customers’ needs, wants, or preferences. Collaborative filtering is a salient technique to support automated recommendations. However, the traditional collaborative filtering approach mainly relies on the assumption that all the given preferences are equally important, irrelevant of when a preference is collected. This assumption ignores the fact that a user’s interests may be changed over time, and the prediction outcome of the traditional collaborative filtering approach may be misguiding if the preferences given at different time are not distinguished appropriately. Therefore, we propose a new collaborative filtering approach to take user interest evolution into account. Specifically, a clustering algorithm is first adopted to group the similar items. Subsequently, for a user, the preference of each cluster is calculated by the given preferences on each item in this cluster as well as the corresponding timestamps. A user’s interest is then represented as a vector containing the preferences of all clusters. As a result, users with the most similar interest vectors to that of the active user will be chosen as his/her neighbors for collaborative recommendation. The experimental results demonstrate that our proposed approach improves the recommendation effectiveness in comparison with the traditional collaborative filtering approach.


Recommendation, Collaborative Filtering, Time Weight, User Interest Evolution, Clustering


ISBN: [978-1-86435-644-1]; Full paper