A Noise-aware Asymmetric Spectral Regularization Collective Matrix Factorization Algorithm for Recommender System in Cloud Services
Juntao Han, Yong Pan · 2020
In cloud computing environment the problem of low recommendation accuracy caused by the influence of social noise is ignored in the existing social recommendation researches, we propose a novel approach to compute the trust values by using the asymmetry of trust network, which can mitigate the impact of social noise. At the same time the overall distribution information captured by different modeling schemes is significantly different when it exists a large amount of data in cloud services, and it affects the user's cold start, as that is not conducive to the personalized recommendation. We study the sparseness problem of the recommender systems, and report the research status of social recommendation. Accordingly we propose a noise-aware asymmetric social collective matrix factorization model called AsySocialCoMF to solve the problem above. The experimental results show that the proposed approach can improve the recommendation accuracy by 2%-15% compared with the current works, and it also performs better for the user's cold start problem. It is worth noting that the approach is also applicable to other information retrieval scenarios.