A hybrid three-way recommendation considering users variability

Hybrid recommender systems leverage diverse information sources and techniques to enhance performance. Nevertheless, integrating users’ multifaceted preferences remains challenging due to the uneven data. Meanwhile, information insufficiency introduces uncertainty in recommendations while existing strategies (i.e., recommend or not-recommend) lack the flexibility to address it. Additionally, these works mainly overlook that ratings not only reflect preferences but imply users’ attitudes toward the strategies, leading to the same recommendation rule despite users distinctly. To solve these issues, a Hybrid Three-Way Recommender (HTWR) system is proposed to formulate personalized three-way rules. Specifically, users’ historical and predictive preferences are captured via tags and ratings while integrated based on the user’s data distribution. Then, the theory of three-way decision is introduced to address such uncertainty by offering the option of defer-recommend. Finally, the users variability is formally given and incorporated into the loss function to obtain personalized rules. Experiments on three public datasets validate the superiority and flexibility of the proposed HTWR.