Deep Confrontation Generation Recommendation Algorithm Based On Group Influence

In the era of large data,the whole information recommendation system has gradually become an important means to obtain valuable information.How to select effective information from a large amount of information that is highly interesting and in line with users' own preferences has become an urgent problem for recommendation system.It is suggested to systematically analyze the historical user behavior and the relationship between potential users and the project,and makes recommendations according to the relationship and user's behavior data.There are still many shortcomings in the traditional recommendation system: there are high sparsity problems in the massive project data,and extracting effective data features becomes one of the problems that traditional recommendation algorithms cannot solve;Compared with the massive project,User behavior data is too short,new users have cold start problem,with the increase of information,based on deep learning Different from tradition: the deep learning recommendation algorithm can better deal with the data with high sparse degree,better extract the advantages of deep data features,Compared with the traditional recommendation algorithm,this algorithm has better accuracy.The content of this paper is as follows:(1)A deep self encoder recommendation algorithm with group influence factor is proposed.In the process of recommendation,the algorithm first clusters the users according to the user's portrait information,and divides the whole user into different user groups.In each user group,the average value is calculated by adding the historical behavior data of all users(except the current user)in the group,and the obtained data is understood as the group influence of the user's group and the current user's history The weighted sum of historical behavior data.The idea of group influence not only acts as noise,provides de-noising function for the training process,but also makes up for the problem of sparse user historical behavior data to a certain extent.In this way,the algorithm has very nice robustness,improves the accuracy of the recommendation algorithm.The experimental results show that the algorithm is 4% higher than VAE and 9%higher than DAE.(2)Deep confrontation generation recommendation algorithm based on group influence.The characteristics of self encoder are that the input data is reconstructed to the maximum extent,but there are still some problems in the process of reconstruction capability and the model is easy to fit.In view of this,this paper uses the idea of the counter generation network for reference,and integrates the idea into the deep self encoder which has added the group influence factors.The ability of reconstruction and anti over fitting of the encoder is further improved by using the idea of the counter generation.After experimental analysis,the algorithm has good performance under the evaluation criteria of accuracy and recall rate.Compared with AVAE and CDAE,this algorithm improves about 4%,37% higher than SLIM and LFM,60% higher than User CF and Item CF.