Gaussian mixture matrix factorization model for robust collaborative recommendation
Cong Li, Zhigang Luo · 2011
Collaborative filtering has emerged as a promising recommendation technique for recommender systems. One of the widely-used techniques for collaborative filtering is matrix factorization. However, it has been noticed that the existing matrix factorization algorithms suffer from unsatisfactory robustness in the presence of rating noises. This paper ascribes the negative impact of rating noises to the assumption that ratings are generated from Gaussian distribution, thus proposes the Gaussian mixture matrix factorization(GMMF) model for robust collaborative recommendation through enhancing the Bayesian probabilistic matrix factorization model by using Gaussian mixture as rating distribution. Experimental results show that GMMF is more resistant to rating noises, and it can effectively improves the predictive accuracy.