Collaborative Filtering Recommendation Method Based on ART2 Dynamic Clustering

Abstract

Based on the extreme sparsity of user rating dat a in collaborative filtering recommendations and the recomme ndation problem for new users, this paper constructs a persona l feature matrix for users by collecting user registration inform ation. It adopts the ART2 neural network dynamic clustering a lgorithm along with the user's personal feature matrix to classi fy users, find neighboring users for the target user, predict rati ngs for unrated items, and improve the response time and accu racy of online recommendations. Experimental results demonst rate that the improved algorithm significantly enhances the rec ommendation quality of the recommendation system, especiall y in situations where user rating data is extremely sparse. Addi tionally, the algorithm effectively addresses the recommendatio n problem for new items.