Dynamic item-based recommendation algorithm with time decay

Nowadays, the customers involved in e-commercial business are increasing rapidly. To meet their needs, many famous companies, like Amazon and Netflix, place building and optimizing e-commerce recommender systems as their priority. Recommender systems aim to provide personalized advice through mining and discovering the interests and consuming patterns of customers. Generally speaking, recommender systems use two strategies to provide recommendations, namely contentbased and collaborative filtering (CF). Furthermore, two primary approaches, namely user-based and item-based, are widely used to build the CF-based top-N recommender systems. Item-based approaches have been empirically proved to provide comparable or even better recommendations than those provided by userbased approaches. In this paper, we first introduce the concept of “time decay” by giving its mathematical definition and redefine the item-to-item similarity function based on time decay. Then we study three patterns of time decay and show their effects on recommendations. Based on the above work, finally we present the dynamic item-based top-N recommendation algorithm that uses time decay to build models and provide recommendations. Our experiments on real data show that the proposed algorithm provides better recommendations.