A Dynamic Prediction Model for Recommender Systems Based on the Doubly Structural Network

In the context of recommender systems, there are two types of enties: users and items, and three types of relationships: users’ relationship, items’ connection and interactions between users and items. In most literatures, one or more of these entities and relationships are used to predict users’ preference or taste. In this paper, we propose a novel approach which incorporates these two entities and three relationships into one framework based on doubly structural network (DSN) . We also develop a dynamic prediction model to learn users’ preference over time by focusing on the active user-item pair’s influence on the corresponding neighborhood. We conduct an experiment and anylize the sensitivity of the model’s parameters and compare the new approach with conventional collaborative filtering (CF) approaches and the results show that the new approach could give a better performance than user-based CF and item-based CF approaches for recommender systems. Key-Words: recommender systems, doubly structural network, expected preference, predictive preference, dynamic prediction model