A graph-based collaborative filtering algorithm combining implicit user preference and explicit time-related feedback

Collaborative filtering is one of the most extensively utilized recommendation algorithms in the e-commerce industry. It typically relies either on implicit or explicit feedback. The existing collaborative approaches fail to capture changes in user preferences while integrating implicit and explicit data. To model the user's current preference, we propose a novel graph-based CWALK algorithm that combines time-related item correlation explicitly and the user's preference for an item implicitly. In the first stage, we cluster users based on their rating behavior, and in the second stage, we combine implicit and explicit feedback to construct a matrix for each user group. A random-walk-with-restart is employed on the matrix to generate a recommendation for each user. Extensive evaluation using the real-world MovieLens dataset shows that the proposed method improves the accuracy of recommendations.