Research On Recommendation System Based On GAN

With the popularization of the Internet and the rapid development of computer technology,all kinds of information resources on the network have also experienced substantial growth.Users often in the face of a large amount of information helpless,do not know how to from the vast amount of information accurate and fast to get relevant information.The user's demand for information can not be satisfied,resulting information resources is wasted and resource utilization is reduced,this is the problem of information overload.An effective way to solve such problems is the recommendation system.In recent years,with the rapid development of artificial intelligence technology,the research of deep learning technology in the field of recommendation system has become one of the hotspots.Among them,the Generative Adversarial Networks(GAN)has made breakthrough progress in the fields of image processing and natural language.The rapid growth of information resources also makes the number of users and projects increase rapidly in the information,resulting in an increasingly small proportion of effective scoring in the dataset and the data is becoming more and more sparse.In addition,it also makes the traditional recommendation algorithm unable to recommend for users accurately.The traditional GAN model simply made use of the scoring relationship between the user and the project,and didn't make full use of the other attribute information of the user and the project,which affected the recommendation accuracy of the algorithm to a certain extent.In view of the above problems,this paper proposes a recommendation algorithm(IRGAN-WD)to fuse the existing IRGAN and Wide & Deep Models,and make full use of the other related attributes of users and projects by using the Wide & Deep model.In addition,the attention mechanism is introduced into the Wide & Deep model,the different weights are dynamically assigned according to the different attributes of different users and projects,the information in the depth is excavated,the identification accuracy of the discriminant retrieval model is improved,and the generation retrieval model is better trained.Experimental results show that compared with other algorithms based on GAN,this method has better recommended prediction accuracy and recommended sorting quality.At present,the number of domestic video and users is growing rapidly,and the traditional film recommendation system has the disadvantages of low data utilization,high system pressure and poor real-time performance.In view of the above problems,this paper deploys Hadoop distributed cluster configuration to ensure the integrity of data,to satisfy the storage of large-capacity data,and uses Spark to carry out distributed computing of the recommended algorithm model,to improve the operating efficiency of the algorithm model,and uses Flume and Kafka to online collect user behavior information.The detailed design realizes the offline recommendation module based on IRGAN-WD algorithm model and the film recommendation system jointly recommended by online recommendation module based on online user behavior data collection and processing.The system test shows that the system has good recommendation accuracy,stability and real-time performance.