Research On Multi-Agent Recommendation Method Based On Deep Learning

At present,with the rapid development of Internet technology,the utilization rate of network has declined dramatically due to the massive information resources.Therefore,how to recommend the information to meet the needs of users,and improve the personalization of the recommendation system and the accuracy of the recommendation effect is currently an outstanding research point to be solved at present.In this paper,the traditional recommendation algorithm is improved to improve the personalized service effect and the accuracy of the recommendation results of the recommendation system,and the deep learning technology and the recommendation system are combined into the multi-agent recommendation framework to carry out in-depth research.The main contents of this paper are as follows:(1)Aiming at the problems of cold start and low accuracy in traditional collaborative filtering based recommendation,a matrix factorization recommendation algorithm based on K-means(KMMF)is proposed on the basis of collaborative filtering algorithm.The algorithm first extracts users with similar interests,and then factorize the matrix based on them.Small scale matrix operation can effectively reduce the computational complexity.What can be found from the experimental results is that compared with the traditional content-based recommendation and matrix factorization recommendation,the improved KMMF algorithm has a significant improvement in recall rate and precision,and the KMMF algorithm has also achieved good results in dealing with the user cold start problem.(2)In view of the lack of considering the interaction of features in traditional recommendation algorithms,a hybrid recommendation model based on deep and cross factorization machine(DCFM)is proposed by integrating deep learning and recommendation system.The model is composed of three parts: factorization,cross network and deep neural network,which can mine features from multiple angles.Compared with the factorization machine(FM),deep factorization machine(DeepFM),deep cross network(DCN)model,the proposed DCFM recommendation model improves the accuracy,balanced F-score(F1 score)and AUC value from the experimental verification on movielens dataset.(3)In view of the interaction between features mined from different perspectives in deep learning technology,the xdeepfm model is studied.The model consists of linear regression unit,fully connected neural network and compressed interaction network(CIN).By comparing XDeepFM model with the single FM and DNN model,the results show that the XDeepFM model which integrates the compressed interactive network can effectively extract the vector level interactive features,which is helpful to improve the accuracy of recommendation.(4)So as to enhance the accuracy,diversity and personalization of recommendation,the recommendation system based on multi-agent is studied.Multi agent recommendation framework integrates DCFM,XDeepFM,which is based on deep learning,and traditional recommendation algorithms,as well as various other types of recommendation.Each recommendation algorithm is encapsulated in an agent.Through the comparison of experimental results,it shows that the recommendation system based on multi-agent can make customers truly experience personalized results.