Research On Collaborative Recommendation Algorithm Based On Probablilistic Matrix Factorization

With the popularity of Internet, electronic commerce is more and more attention, the information in the network increases gradually, even the information overload problem, in order to solve the problem of information overload, recommendation system arises at the historic moment. However, due to more and more data in the network, only a handful of user rating will be on the same project, there are a lot of projects no users to its score, data sparseness is becoming more and more serious, so the data precision is also a gradual decline in the recommendation system. In order to alleviate the data sparseness problem, people have to look for other resources, trust network is a resource that can be used, but trust network also face the data sparseness problem. Aiming at these problems, how to alleviate data sparseness, improve the prediction accuracy of recommendation system became the hot problems of attention. In this paper, on the basis of the research at home and abroad, this paper carried on the thorough research for the problem.Firstly, to solve the problem of data sparsity, we propose a novel collaborative recommendation algorithm which integrates the probabilistic matrix factorization and neighbor models. We first propose a method to calculate the similarity between users or items based on the probabilistic matrix factorization model and construct a natural exponential function to compute the weighted similarity. Then we devise a collaborative recommendation algorithm to make recommendations for the target user, which dynamically adjusts the recommendation results for user- and item-based models by the balance adjustment factor.Secondly, we propose a novel collaborative recommendation algorithm which integrates the probabilistic matrix factorization and trust model. We also used the probabilistic matrix factorization model to calculate the similarity between users. Then, we compute the set of similar users based on the trust matrix, converting the similarity, and predict the potential trust relationship; Fused the similarity and trust, to calculate the weights between the user, then used the weight instead of the similarity in user-based neighbor model algorithm, implement collaborative recommendation.Finally, we compare the experimental evaluations and analysis of the algorithms proposed in this paper with the existing traditional methods.