A Research About Probability Matrix Factorization Recommendation Algorithm Based On Community Matrix

Under the era of big data,the amount of Internet information has exploded,and it is difficult for users to find the information resources that individuals need in a huge amount of information.This is the problem of information overload that can't be avoided in the future.As the important part of the recommendation system,the recommendation algorithm can solve the problem of information overload.It's meaningful to study the recommendation algorithm to improve the performance of the recommendation system and expand the recommendation system function.As a branch of the collaborative filtering algorithm,the probability matrix factorization algorithm has profound influence in the industry due to its excellent characteristics.However,the probability matrix factorization algorithm is also affected by the cold start problem.In order to improve the accuracy of the recommendation algorithm and solve the cold start problem,this paper uses the community to process the user's social information from the perspective of user's cluster collaboration.Then,combining the community and the probability matrix factorization algorithm,this paper proposes a probability matrix factorization recommendation algorithm based on community matrix(CPMF).To make the best level of aggregation for the users,the paper proposes a SLPA community division algorithm based on modularity(QSLPA),and the algorithm divides the social network composed of user's social information into target communities with the largest modularity.According to the affiliation between the user and the community,the paper defines the community matrix that describes the user's social information in the form of matrix.Integrating the community matrix and the user rating matrix composed of user rating information,we utilize the probabilistic matrix factorization method to extract implicit eigenvector such as user,item,user community,and community.Considering the influence of user community feature on user feature,we propose the community adjustment parameter,and derive the objective function with eigenvectors.In order to make the prediction rating close to the actual rating,the paper uses the stochastic gradient descent method to minimize the objective function.The related implicit eigenvectors are optimized to the optimal solution by iterative learning,and we can predict the rating by multiplying the implicit user eigenvector and the implicit item eigenvector.In order to ensure the generality of the experiment,our experiments choose the two public datasets of Last.FM and Epinions,and select the recommendation algorithms such as PMF,SoRec,TrustMF,TrustPMF as the comparison object.We take the mean absolute error(MAE)and the root mean square error(RMSE)as experimental indicators to measure the recommendation performance of the algorithm.Through the comprehensive experiments on the two datasets,all the experimental results prove that the CPMF algorithm has good recommendation accuracy,and the CPMF also has good performance on the cold start problem,and the CPMF algorithm's advantage in root mean square error is more obvious.All the advantages are attributed to the cluster cooperation of users in the community.