A novel web recommender system considering users' need evolution

The recommender systems' task is to predict items in which a user might like in future. Typically they do it in two stages. One, offline stage, that extracts patterns from mining users' historical behavior, and another, online stage predicts items by matching the active requests with the patterns; but without considering other interests of the active user. In general, web users' navigations contain gradual evolution of their needs. We conducted an approach that addresses the evolving nature of users' needs and interests in making recommendations. Since users' behaviors are not stable and their interest changes with time, our approach is based on interest-drifting. One of the main shortcomings of interest-drifting based approaches is that they assume different users have the similar interest changes over time and do not consider this important factor that the trend of interest changes for users is different. This paper attempts to tackle this problem, too. Up to now there has not been any study to tackle these problems. We evaluate our proposed methodology on EachMovie dataset. Results show a better performance than existing approaches.