Research On User Reputation Collaborative Recommendation Algorithm Based On Three-way Decision

Due to the rapid development of Internet communication technology,people in the 21 st century are in an era of information explosion.Facing the vast amount of information and data at the present stage,it is not easy for users to find the information they need.In this context,the recommendation system was born.Recommendation system is an effective way to replace search algorithm,which can help users find some items they may be interested in but cannot find.As one of the most classical types of recommendation algorithms,collaborative filtering algorithm is popular in academia and industry,and is widely used.As a typical recommendation algorithm,collaborative filtering algorithm still faces many problems,such as scalability,accuracy,cold start,sparsity and so on.Due to the sparsity of the data,the information obtained is insufficient,so the recommendation accuracy is insufficient.This paper makes an in-depth research on the existing collaborative recommendation algorithms at home and abroad,focusing on how to improve the recommendation accuracy of collaborative recommendation algorithm,and some problems such as the potential characteristics of users in social networks.Main work of this paper:1.Firstly,this paper briefly introduces the theoretical knowledge of collaborative filtering recommendation algorithm,reputation system and three decision making,and investigates the research status and existing problems of collaborative filtering algorithm and reputation-based recommendation algorithm at home and abroad.Then the user reputation collaborative recommendation algorithm based on matrix decomposition technology is introduced in detail.Finally,aiming at the problem of insufficient accuracy in the current collaborative recommendation algorithm of user reputation,two effective collaborative recommendation algorithms based on fairness and tolerance are proposed by combining the user reputation system with the collaborative filtering recommendation system.2.Collaborative filtering algorithm is a recommendation technology widely used in e-commerce websites.However,due to sparse data,the information obtained is insufficient,so the recommended precision is insufficient.The user reputation system can be established by analyzing user rating data and making full use of user reputation to supplement information,which is helpful to improve the accuracy of recommendation.In this project,a three-way User Reputation Collaborative Recommendation Algorithm Based on Fairness(TWDA)is proposed.First according to the user's evaluation data for the user fairness,based on the ideaof three decisions that will be a fair score is defined as the domain(POS),unfair score(NEG),defined as negative domain defines indistinguishable score as edge boundaries(BND),using edge boundaries parameters reasonably allocate the score boundary domain to the positive domain or negative domain,then based on Beta distribution computing user reputation,then the user reputation system combined with the matrix decomposition model.In the experiment,MAE was used as the evaluation index.The results of the two classical data sets,movielens-100 k and movielens-1m,showed that compared with the existing traditional recommendation algorithm,TWDA algorithm was effectively improved in accuracy.3.The existing user reputation calculation methods seldom consider the impact of user tolerance on their reputation.Therefore,this paper focuses on the user reputation collaborative filtering recommendation algorithm based on tolerance.A three-way User Reputation Collaborative Recommendation Algorithm Based on Leniency(URBL)is proposed.First,the user's tolerance is calculated based on the user's rating data.Then,users are divided into three categories according to the tolerance: demanding users,ordinary users and relaxed users.Based on the thought of three decisions,strict users are defined as positive field(POS),loose users as negative field(NEG),and general users as boundary field(BND).For users in different regions,different methods are adopted to judge whether the rating given by users is fair.Then the user reputation is calculated based on the Beta distribution,and the user reputation system is combined with the matrix decomposition model.In the experiment,MAE was used as the evaluation index.The results of the two classical data sets,movielens-100 k and movielens-1m,showed that the accuracy of URBL algorithm was effectively improved compared with the existing traditional recommendation algorithm.