Research And Implementation Of Recommendation Algorithm Based On Deep Reinforcement Learning

In the context of the era of big data,the information explosion and data redundancy problems annoy many users who desire to obtain efficient and accurate information,and also hinder the development of various industries.Especially for content recommendation platforms,users cannot quickly find their own items of interest in a lot of information.Advances in the technology of recommendation systems have broken this dilemma.Personalized recommendation is achieved through algorithms such as data mining and deep feature extraction.It puts aside redundant information and hits user interests.Today,it can be said that it is an operation platform of major companies essential technology.Although the development of the current recommendation system has greatly accelerated social progress,it also faces various difficult problems in different application scenarios.For example,the cold start of user information and content information and the sparseness of data pose challenges to technology implementation,and the diversity of different content characteristics and user interest changes affects the system recommendation effect and real-time performance.In addition,the scalability of the system in the face of data overload will affect the operation of the platform.This paper focuses on the cold start,data sparsity,and scalability problems of traditional recommendation algorithms,and makes the following studies:(1)A reinforcement learning recommendation algorithm based on deep feature extraction is proposed.The movie recommendation is used as an application scenario for algorithm verification.According to the data type,we perform data analysis and set relevant states,actions,reward values,and deep reinforcement learning networks to model the complete recommendation process as a deep reinforcement learning model.Based on this,in order to allow the system to capture user interest changes in real time,a prioritized experience replay mechanism adapted to the system is proposed.Experimental results verify that the model is superior to traditional collaborative filtering recommendation algorithms in recommendation accuracy.(2)Considering the continuous action application scenario and the problem of overestimation and long training time of the model itself,an optimization model DPGQN is proposed.In the experimental verification phase,the two systems proposed before and after are compared.The theoretical conjecture was verified.DPGQN shortens the training time and improves the accuracy of the system recommendation.