Research On Recommendation Algorithm Based On Popularity And Internal Similarity Of Users

Recommender systems have been widely used in many fields such as e-commerce,movies,and music.As the basis of recommender systems,recommendation algorithms are a very important research topic.Although researchers have proposed a variety of improved algorithms,the current recommendation algorithms still face problems such as low accuracy,cold start,and vulnerability to attacks.In terms of the questions,this thesis proposes several improvements,focuses on improving the accuracy and robustness of the recommendation algorithm.The specific research work is as follows:(1)In order to solve the problem of unreasonable allocation of resources by traditional mass diffusion and heat conduction algorithms,this thesis proposes a dynamic resource allocation algorithm based on popularity.Previous studies have shown that the popularity of items plays an important role in the recommender system.Considering the influence of the popularity of items in different periods on the recommender system,this thesis improves the resource allocation process of the mass diffusion algorithm and heat conduction algorithm,and proposes a novel unbalanced resource allocation algorithm.The experimental results prove that compared with the mass diffusion algorithm and heat conduction algorithm and some improved algorithms based on mass diffusion and heat conduction,the proposed algorithm can significantly improve the performance of the recommender system.(2)In terms of the problem that the recommender system is vulnerable to cyber naval attacks,this thesis proposes a method based on internal similarity of users to improve the robustness of the recommendation algorithm.Recommender systems should not only focus on accuracy and diversity,but also improve robustness to respond to shilling attack.However,existing research has paid little attention to this point.This thesis systematically studies the influence of different water army behaviors on the recommender system,and finds that false and redundant information in the system will reduce the accuracy of the recommender system.This thesis proposes an improved KNN algorithm by merging internal similarities between users in user-based collaborative filtering and material diffusion algorithms.The experimental results show that the proposed algorithm can reduce the influence of the information noise inthe system on the performance of the recommender system to a certain extent,and improve the robustness of the recommender system.