Collaborative Filtering Based on Clustering and Simulated Annealing
Wei Guo Song, Siqi Liu · 2021
Collaborative filtering (CF) is a typical and widely used recommendation method that can be categorized into user-based CF and item-based CF. For user-based CF, how the approach chosen more suitable similar users is important. To solve this problem, we first define the user-genre vector as the objects used for clustering. The user-genre vector uses not only genre information to enrich the source data for recommendation but also users’ rating biases, which can be differentiated, to certain extent. Then, we propose a k-means clustering method enhanced by simulated annealing to ensure stable clustering results. Finally, we define comprehensive similarity, which combines original and transformed ratings, to select more appropriate similar users for recommendation. Our experimental results showed that the proposed method can make recommendations accurately and reliably.