Multi-granularity sequential three-way recommendation based on collaborative deep learning

Recommender system (RS) is an information processing system, which mainly utilizes the recommendation information (RI) learned from different data sources to capture user's preference and make recommendation. However, existing recommendation strategies primarily focus on the static recommendation strategy, and the multilevel characteristic of RI is ignored. To address the above-mentioned problem, we introduce the idea of granular computing and sequential three-way decisions into RS, and then propose a naive recommendation method with cost-sensitive sequential three-way recommendation (CS3WR) based on collaborative deep learning (CDL). Firstly, inspired by the structure thinking of granular computing, we design a CDL-based joint granulation model to produce the multilevel RI. Subsequently, we propose a CS3WR strategy and an optimal granularity selection mechanism to get the optimal recommendation and optimal granularity, respectively. Finally, extensive experimental results on two CiteUlike datasets validate the feasibility and effectiveness of our methods.