Research And Implementation Of Personalized Recommendation System

The rapid development of Internet technology has turned the world into a sea of information mass convergence.People can easily get a large amount of information and enjoy the convenience of the Internet era.But because of the large and complex information,people are also facing more and more meaningless choices while acquiring information.In order to reduce the burden of users in the face of cumbersome choices,recommendation algorithm emerges.It is committed to recommend information that is more in line with their preferences and provide users with more accurate and convenient information services.At present,most of the recommendation algorithms adopt fixed strategy.Although it has high recommendation accuracy,it lacks diversity and flexibility.The recommendation results have the disadvantages of high similarity,so it is difficult to perceive the user’s immediate change preference,and can not give full play to the value of personalized recommendation system for a long time.Therefore,this paper will focus on the research and design of a personalized recommendation system which takes into account accuracy and diversity.To provide more intelligent information recommendation services for users,this system not only can accurately recommend information for users,but also will improve the system’s sensitivity,and make timely response according to the user’s preference changes.Firstly,to solve the problem of low flexibility of recommendation system,a recommendation algorithm based on reinforcement learning is proposed.By simulating the user and recommendation process into the agent and environment in reinforcement learning,building a framework of reinforcement learning based on Markov,FM model is used to extract cross features.Combined with the depth neural network as the approximate value function in Q function,and the optimal strategy is determined by learning the maximum expected value function.In order to avoid the problem of data correlation and local optimization,we can select random data from experience pool to strengthen training by experience playback,so as to optimize network parameters and improve the flexibility of the system.In the process of recommendation,the recommendation algorithm based on reinforcement learning makes full use of the dynamic interest change factors of current users,which can effectively improve the overall diversity of the final recommendation results.Then,a collaborative filtering method is proposed to solve the problem that the recommendation list is not accurate in short time.The value function estimated in the intensive learning is taken as the measurement standard for calculating user similarity in collaborative filtering.At the same time,user rating and user evaluation are added in the calculation of value function to better reflect the division between different users,fully mining the user’s personalized interest characteristics.It will make more accurate in computing user similarity,and improve the accuracy of system recommendation Degree.In this paper,we combine two methods to propose a hybrid recommendation algorithm based on reinforcement learning.Compared with several current collaborative filtering recommendation algorithms,it is verified that the hybrid recommendation algorithm based on reinforcement learning has better accuracy,and the overall diversity of recommendation results is higher when the recommendation list is longer.The proposed hybrid recommendation algorithm based on reinforcement learning can ensure the accuracy of recommendation,and its recommendation results are also of high diversity.It fully reflects the value of personalized recommendation algorithm,this work is expected to further improves the practical value of it.