Research On Multi-relation-based Matrix Factorization Recommendation Algorithm

The rapid development of the information age has provided people with a lot of convenience,but a large amount of data has also brought people a certain amount of trouble,making it difficult for users to choose information.Therefore,the emergence of recommendation systems is to help people solve this problem.Since every user exists in various social relationships,combining the social relationship with the recommendation system can effectively help people get the information they want.In the real world,a user cannot exist in a social relationship,and for the user,and the effect of each social relationship on the recommendation effect is different,the existing social recommendation algorithms often only introduce A social relationship,so it will greatly reduce the accuracy of the recommendation.In order to improve this phenomenon,this article is based on the existing social recommendation algorithm to realize the combination of the recommendation system and various social relationships,and the results are obtained through experiments.The specific research content and results are as follows:1.The traditional matrix factorization recommendation algorithm does not introduce social relationships,or only introduces one social relationship,ignoring the effect of multiple social relationships on the recommendation.Therefore,this paper proposes MDRS2 recommendation based on the multi-subnet complex network model The algorithm first decomposes the user product rating matrix to obtain the potential feature space of the user and the product,performs inner product through these two feature spaces,and then predicts the blank score,obtains the loss function according to the proposed model,and then uses the gradient descent to continuously Iterate until the optimal solution is obtained,and use the obtained predicted value to complete the recommendation to the user.The experimental results show that the recommendation algorithm that combines multiple social relationships improves the recommendation result compared to the previous recommendation algorithm that does not add social relationships or only considers a single relationship,so it can better serve users.2.Although the introduction of social relationships improves the recommendation results,for the entire data set,the data still has a high sparseness.Therefore,a probabilistic matrix decomposition recommendation algorithm based on community structure division is proposed,and the clustering algorithm is used to reduce the influence Users with similar similarities are divided into the same community in order to improve the sparseness of the data.Then according to the collaborative filtering recommendation algorithm,the neighbor users of the user are found according to the conditional probability,and the blank score is predicted through the scores given by the neighbor users.Through this step After completing the first data filling,a brand new user product rating matrix will be obtained.Combined with the MDRS2 recommendation algorithm model proposed above,the filled data will be decomposed again to complete the second data prediction,which will greatly reduce the sparseness of the data.The experimental results show that the recommendation results are further improved by reducing the sparsity of the data.