Dynamic evolutionary clustering approach based on time weight and latent attributes for collaborative filtering recommendation

Abstract Collaborative filtering is one of the most widely used individual recommendation algorithms. The traditional collaborative filtering recommendation algorithm takes less care of time variation, which may be inaccurate in real surroundings. A novel dynamic evolutionary clustering algorithm based on time weight and latent attributes is proposed. According to the time effect of historical information in recommendation system, forgetting curve is introduced to better grasp the recent interest of the users. To gather users with similar interest into the same cluster, item characteristics and user attributes are mined. Therefore, network model is established by introducing the forgetting function to score matrix, utilizing item characteristics and user attributes. Items and users are regarded as heterogenous nodes in network. Furthermore, a novel dynamic evolutionary clustering algorithm is adopted to divide users and items set into K clusters, and individuals with higher similarity are clustered. The preferences of users in the same cluster are similar. Then, collaborative filtering is applied in each cluster to predict the ratings. Finally, the target users are recommended predicted according to prediction ratings. Simulations show that the presented method gains better recommendation accuracy in comparison of existing algorithms through MovieLens100k, Restaurant & consumer and CiaoDVD data sets.