Most recommender systems use collaborative filtering to predict new items of interest for a user. In this paper we present a model-based approach to collaborative filtering by using unsupervised self-organising ART2 neural networks which deploys two of the main advantages of the ART model - stability and plasticity when the system works in poorly defined domains and planning of network resources is difficult or even impossible We report empirical results that show the impact of ART2 NN parameters on recognition stability, appropriate category granularity, classification accuracy, and response time.
展开▼
