An Improved Collaborative Recommender System

Collaborative Filtering is widely used in the recommendation system which gives recommendation based on the similar liking users. The Collaborative Recommender System (CR) suffers from sparsity and cold-start problems. We design and implementation an intelligent kernel based Fuzzy Collaborative Recommender System for map user needs to products that can satisfy them based on ratings. In this paper we present an intelligent, smoothened, fuzzy kernel based approach to solving this problem. The system is intelligent, so the user gets recommendation based on fuzzy similarity with the neighbours. The approach is smoothened, so it uses radial basis function network to smooth the sparse rating matrix. In this manner, it allows the efficient determination of products that meet the user´s requirements. Experimental results based on the Movie lens dataset show that the proposed system only by smoothening provides more than 95% accuracy for existing users in the highly sparse data. The results proved that the proposed system outperforms existing systems such as Support Vector Machine, Multi Layer Perceptron, Bayesian, and Singular Value Decomposition, in terms of accuracy and relevance of recommendations. Finally the system is developed for our college library book application and it is tested by our college students.