Collaborative Filtering Recommendation Model Based On Probability Matrix Factorization And Genetic Algorithm

Nowadays people are faced with the problem of “information overload”,and in other respects,the requirements of different people are different.Commercial companies generally adopt recommendation systems to meet people's individual requirements.Collaborative filtering is one of the most recommended algorithms currently used by commercial companies.However,in the process of applying the collaborative filtering algorithm,there will be problems of sparseness,cold start,and parameter tuning in a variety of application scenarios.In order to solve these problems,this paper constructs a new recommendation model based on collaborative filtering.The specific work is as follows:(1)Using the probability matrix factorization model to solve the data sparse problem.Firstly,the initial user similarity and the neighbors of each user are obtained from the original data.The probability matrix factorization algorithm is used to fill the scoring matrix,and the similarity and the optimized neighbor selection set are corrected based on the filled data,and the scores of the target items by the neighbors are obtained.To estimate the current user's rating of the project,so that the problem of data sparseness can be solved to some extent.(2)Using the similarity algorithm based on item tag matching to solve the cold start problem.When calculating the project and project similarity problem,in addition to using the user's ratings for the project as the basis,the project tag is added as the basis for the similarity calculation,and given a certain weight,for the common cold start problem in the recommendation system.Played a certain help.(3)Using the related techniques of genetic algorithm and combine it with the model to solve the problem of system parameter adaptation.There are two main points in the innovation of this model:(1)Set the user evaluation quantity coefficient to accurately control the new item from cold to hot,and the recommended results are more accurate.(2)Extract various parameters in the model itself or in combination with the model,optimize the values of these parameters through genetic algorithm,so that the model can self-learn and evolve to achieve the best under various business scenarios and various data sets.For the effect of the model,this paper aims at the common collaborative filtering algorithm(divided into item-based,user-based),generalized probability matrix factorization collaborative filtering algorithm,without genetic algorithm model and the whole model algorithm for comparison,at the same time verification is performed on the public data set movielens and some e-commerce data set.Finally,according to several commonly used recommendation algorithm evaluation criteria,the experimental results and data are analyzed.the results proposed model can solve the data sparseness,cold start and optimal parameters in various business scenarios to some extent.It shows good results on both data sets.