Research On The Recommendation Method Of Deep Reinforcement Learning With Negative Feedback

With the rapid development of user recommendation field,all kinds of technical algorithms such as collaborative filtering,machine learning,and deep learning are applied in recommendation methods.However,traditional recommendation methods have the characteristics of static recommendation algorithm,they can not respond to the dynamic behavior of users,and can only achieve local optimization but not global optimization.However,recommendation based on traditional reinforcement learning algorithms,such as valuebased reinforceme-nt learning,will incur excessive memory and time consumption when they are faced with recommendation tasks of large scale state and behavior dimension.To solve the above problems,and puts forward a kind of based on Reinfo-rcement Learning With Negative Feedback depth recommendation method(WN DR),use with the GRU,helped model of feature extraction to extract the use-r characteristics,rarely pay attention to the general method of using Negative Feedback behavior,so as to improve the characteristics of user modeling,make the model more accurate,At the same time,improved Deep Deterministic Pol-icy Gradient(DDPG)model is used to complete the training of recommendation methods,better handle the dynamic changes of user characteristics,solve the problem of slow convergence of neural network in traditional methods,and improve the training efficiency of recommendation methods.The commodity recommendation method based on deep reinforcement learn-ing with positive and negative feedback was implemented on the real e-comm--erce data set with negative feedback behavior,and the offline training environment of reinforcement learning was established to complete relevant experiment-s,and the results were compared with those of other recommendation methods based on traditional methods.The experimental results show that under the sam-e experimental environment,the proposed recommendation method can be stronger than the traditional recommendation method in some situations,and can improve the training speed of the recommendation method to some extent,and improve the computational efficiency of recommendation method.