Research On Parallel Recommendation Algorithm Based On Trust Mechanism

With the rapid development of Internet,mankind have step into the era of big data.We have to face the problem of information overload while enjoying the benefits of e-commerce.Therefore,recommendation system arises at the historic moment.It can predict users' interests and match items for them according to their data of historical behavior.It saves a lot of time and energy for users and brings much profit for merchants frequently.Many domestic and foreign scholars have proposed kinds of recommendation algorithms which have taken good effect.Collaborative filtering algorithm is the most widely used and the best performed recommendation algorithm.However,there still are some problems limit the development of recommendation system,the most remarkable issues are data sparseness and real time.In response to the problems,this article has done some relevant research and improvement work as follows:(1)In order to solve the problem of data sparseness,User-Trust matrix decomposition(UTMF Model)is proposed,it is a kind of collaborative filtering algorithms based on model.This model has a good performance on predicting default value of user-item score matrix.And this article introduces users' trust relationship to improve the accuracy of recommendation.From the result of experiments,no matter it is based on NF5M dataset or ML1M dataset,we can see UTMF model has better accuracy than regularization matrix decomposition model(RMF Model).(2)In order to improve the problem of real time,A new parallel model(PTSGD Model)is proposed to realize the parallel of UTMF Model.It is inspired by alternating least squares algorithm.It uses stochastic gradient descent algorithm to iterate and realizes parallel by separating users' character matrix from items'character matrix.The model has been proved have a good performance on running time.This article compares it with Distributed Stochastic Gradient Descent Model(DSGD Model)and Alternating Least Squares Model(ALS Model).We find it has better speedup and parallel performance.