Collaborative Filtering Recommendation Based On Probabilistic Matrix Factorization

With the development of mobile internet,the amount of data information increases explosively.Excavating the information that users are interested in is the key factor to improve the stickiness of users.Designing an appropriate recommendation system becomes more and more important in this background.Collaborative filtering plays an important role in recommendation,but many collaborative filtering systems only use the users' rating information to recommend items.It's difficult for us to excavate too much information from it.With the development of network and the popularization of smart phones,the channels of obtaining information have become diverse,and the information of users and items that can be collected has increased.The use of various auxiliary information has become a hot research topic.This paper proposes a recommendation method based on probabilistic matrix factorization,which is a collaborative filtering method using composite information.on the one hand,introducing trust information to constrain user potential vector.The existence of trust relationship can make up for the deficiency of scoring data,and it is also a key factor in the recommendation process.On the other hand,a convolutional neural network is used to extract features from descriptive documents of each item to construct potential feature vectors.The use of convolution neural network can help us to understand document information more deeply,especially those items which lack scoring information and can only rely on describing documents.This algorithm is based on probability matrix decomposition algorithm.Based on the probability matrix decomposition algorithm,by learning rating information,trust information and description document information,the algorithm obtains the potential eigenvectors of users and items from different perspectives and integrate into a unified framework.Through reasonable premises,the algorithm is optimized from the perspective of probability model.Using the main idea of matrix decomposing to predict missing scores.This algorithm alleviates data sparsity and improves the accuracy of recommendation.The proposed collaborative filtering algorithm based on probabilistic matrix factorization is compared with the classical or recently proposed similar algorithm.The experimental results on four standard recommendation datasets show that the proposed method outperforms or closes to the comparison algorithm in most cases.In addition,the collaborative filtering method in this paper can effectively improve the accuracy of recommendation in the case of sparse data.