Research On Matrix Factorization-based Recommendation Algorithm Via Incorporating Multi-information

Personalized recommendation is ubiquitous and has been extensively used in many online services such as e-commerce,advertising,social media,etc..It provide users with the items that are potentially of interest and thus is useful to overcome information overload problem.The core idea is to estimate the possibility of a user buying an item based on historical interactive information such as purchase and click.In matrix factorization based algorithm,user behavior matrix is factorized as the product of user and item feature matrices in the latent factor space,because of its good characteristics of flexibility and scalability and therefore become an important basis for researchers to construct social recommendation models.The recommendation algorithms exist the data sparse problem as well as cold start and scalability issues.Based on the research results of existing collaborative filtering algorithms and social network recommendation algorithms,we propose the research on matrix factorization-based recommendation algorithm via incorporating multi-information.The main contributions are summarized as follows:Firstly,a collaborative filtering recommendation algorithm based on bipartite graph partitioning co-clustering is proposed.Users and items are constructed as a bipartite graph for co-clustering and they are mapped into low-dimensional feature spaces.Then,two improved similarity calculation strategies based on the clustering results,cluster preference similarity and rating similarity are combined.The user-based and item-based approaches are later adopted to obtain the predictions for an unknown rating and finally fuses these resultant predictions.Secondly,a matrix factorization recommendation algorithm fusing reliability and influence propagation is proposed.The existing matrix factorization recommendation algorithm has the problems of ignoring the user recommendation accuracy and failing to consider the degree of trust and the influence propagation between users.The reliability relationship among users is calculated to measure the recommendation accuracy with rating data.Secondly,a reasonable user shared feature space is designed for rating and reliability matrices based on matrix factorization technology.Finally,a reasonable objective function is designed to consider the influence propagation relationship among users in social network,which can effectively solve the problem of cold start.A matrix factorization recommendation algorithm is proposed,which combines reliability and influence propagation.The validations against real-world dataset show that the proposed method performs better than state-of-the-art recommendation algorithms.