A Robust Collaborative Recommendation Algorithm Incorporating Trustworthy Neighborhood Model

Abstract

The conventional collaborative recommendation algorithms are quite vulnerable to user profile injection attacks. To solve this problem, in this paper we propose a robust collaborative recommendation algorithm incorporating trustworthy neighborhood model. Firstly, we present a method to calculate the users’ degree of suspicion based on the user-item ratings data using the theory of entropy and the idea of density-based local outlier factor. Based on it, we measure the user’s trust attributes from different angles by introducing the source credibility theory and propose a multidimensional trust model incorporating users’ degree of suspicion. Then we propose a trustworthy neighborhood model by combining the baseline estimate approach with the multidimensional trust model. Finally, we devise a robust collaborative recommendation algorithm to provide more accurate recommendation for the target user by integrating the M-estimator based matrix factorization approach and the trustworthy neighborhood model. Experimental results on the MovieLens dataset show that the proposed algorithm has better robustness in comparison with the existing collaborative recommendation algorithms.