A local latent space recommendation method fusing social information.

With the development of social networks, more and more studies utilize social information to improve the performance of traditional recommendation algorithms. However, most of the existing recommendation algorithms ignore the diversity of user interests and do not consider the aspects that users care about in different social dimensions, resulting in poor recommendation quality. In order to solve this problem, a recommendation method that considers both global latent factors and the specific latent factors of different subsets is proposed. The recommendation process considers both the user's shared preferences and the user's specific preferences in different subsets. This method first divides users into different subsets according to their social relationships, based on the intuition that users participate in different social dimensions, and are interested in different items. Secondly, the truncated singular value decomposition technique is used to model the user s rating of items, among which the global factors capture levels shared by users, while specific latent factors of different user subsets capture specific levels of user concern. Finally, global and local latent factors are combined to predict user ratings for unscored i-tems. Experiments prove that the method is feasible and effective. [ABSTRACT FROM AUTHOR]