| The rapid development of Internet applications has brought the explosive growth of information and data.In this context,the recommendation system was proposed,which helped users to obtain effective information in massive amounts of data,and gradually became one of the important tools to alleviate the problem of information overload.The key of recommendation system is personalized recommendation algorithm,according to the difference of users' background information and behavior information,personalized recommendation algorithms generate different results,thus satisfying the personalized requirements of the users.When user's demands are not clear,personalized recommendation algorithms can take the initiative to locate user's demands and provide special customized recommendation service.Although personalized recommendation technology is developing rapidly,there are still many problems to be solved,such as poor real-time performance,sparse data and poor scalability.This dissertation respectively introduces the algorithms based on collaborative filtering,content-based and hybrid recommendation algorithm,In this dissertation,we analyze the recommendation algorithm based on Bandit model and probabilistic matrix factorization model,and put forward different improvements.The main research contents are as follows:(1)This dissertation briefly reviews the background and significance of the research on personalized recommendation technology,and enumerates the current research status of various recommendation algorithms,summarizes the formal definition of recommendation system,introduces the recommended algorithms of different categories in detail,analysis of the various types of recommendation algorithms principle and advantages and disadvantages,and provide a theoretical basis for the following research.(2)The traditional based on the Bandit model recommendation algorithm is analyzed.It is found that this kind of algorithm is prone to Matthew effect and long tail phenomenon.A multi-objective optimization recommendation algorithm is proposed for this problem.The algorithm can effectively avoid Matthew effect on the basis of ensuring the accuracy of prediction,and improve the mining ability of the recommendation system to the long tail items.This dissertation uses YaHoo's news recommendation datasets to experiment and evaluate the algorithm.The experimental results show that the multi-objective optimization recommendation algorithm can effectively solve the long-tailed product excavation problem,avoid the Matthew effect and improve the accuracy and the breadth of recommendation system.(3)The traditional algorithms which based on probabilistic matrix factorization model is analyzed.It is found that the proposed algorithm does not consider both the user attribute information and the correlation information between items,and the information will also affect the recommendation result.A new collaborative filteringbased recommendation algorithm is proposed for this problem.The algorithm applies the user attribute information and the correlation information between recommendation objects to the probability matrix decomposition model.First of all,the algorithm mining user attribute information and correlation information between items,and then get the user attributes and the relationship between the items into the basic probability matrix factorization model.MovieLens datasets are used to verify the algorithm,the experimental results show that the algorithm is superior to the existing several algorithms. |
