Research On Personalized Recommendation System Based On Latent Factor Model

Along with the rapid development of Internet technology,people have unprecedented changes in the way they obtain information.As long as they have the corresponding conditions,they can easily obtain rich resource information from the Internet,which brings convenience to users and faces new problem is that the "information overload" problem arises.However,the emergence of the recommendation system can effectively solve such problems,and it can actively recommend the preferred data for the user based on the historical behavior data.The latent factor model is a relatively young recommendation algorithm in the recommendation system.At present,it has been applied in many fields,but there are still some defects in the practical application of this model that are worthy of further study.First of all,the traditional latent factor model recommendation algorithm has a cold start problem and the user history score data is too sparse,which leads to the decrease of recommendation accuracy.Secondly,as the scale of users and items continues to increase,the amount of data that the recommendation system needs to process becomes larger and larger,making the traditional latent factor model recommendation algorithm face a problem of poor scalability.Because the parameters in the model recommendation algorithm are continuously iterated and updated,they are relatively large in terms of time consumption and calculation amount.In order to solve the problems existing in the traditional latent factor model recommendation algorithm,this paper makes a corresponding improvement on the basis of this model,effectively solving the problem of the prediction accuracy caused by cold start and data sparsity.In order to make the improved latent factor model recommendation algorithm have better scalability,this paper solves this problem by parallelizing the recommendation algorithm in a distributed environment.The main research work of this paper is as follows:Firstly,Aiming at the problem that the traditional latent factor model has a cold start and the prediction accuracy is reduced in the face of data sparsity,the model is improved accordingly.By merging the user's own feature attribute information into the traditional model,when the user's score data is sparse or extremely sparse,the user's own feature attribute can be used to obtain the score data of the neighbor user,and then based on the score data of the neighbor user.Make the corresponding recommendation results for this user.Finally,the experimental results verify that the improved algorithm effectively solves the problem of reduced prediction accuracy caused by data sparsity.Secondly,In order to make the improved latent factor model recommendation algorithm have better expansibility when processing large amounts of data,this paper proposes a solution to parallelize the recommendation algorithm in spark distributed environment.In order to verify whether the improved recommendation algorithm has good scalability in the face of a large amount of data in a distributed environment,experiments were carried out on MovieLens data sets of different sizes,and the experimental results verified that the improved recommendation algorithm has good scalability in spark distributed environment operation.Thirdly,Based on the improved latent factor model in this paper,I designed a prototype recommendation system based on spark big data platform,and finally completed relevant functional tests on the front-end interface of the prototype system.