Research On Personalized Recommendation Algorithm Based On User Characteristics

With the rapid development of Web2.0,the number of Internet users has reached hundreds of millions of scales.Different users have different user feature.How to recommend information that users are interested in or that users may potentially need,according to user feature in the numerous complicated resources of the Internet.Obviously,this become a hot issue in the present study.This paper is based on this as a starting point,from the perspective of user feature,relevant research on the recommendation algorithm.Firstly,according to the time effect feature of user's preference for resources selection,this paper proposes a recommendation algorithm fusing mass diffusion heat conduction and time effect.Secondly,this paper proposes a new method for the user's label feature and the importance of individual users,that is a recommendation algorithm based on the multi-label clustering and core user;finally,based on the above two algorithms as the theoretical basis,a news recommendation prototype system is designed and implemented.The specific work studied in this paper is as follows:(1)Researched the time-effect feature of users on resource selection,and proposed a recommendation algorithm fusing mass diffusion heat conduction and time effect.Firstly,the mass diffusion recommendation algorithm and the heat conduction recommendation algorithm based on the bipartite network will be used mixedly.Then,on the basis of this,the user's time-effect feature of resource selection will be analyzed,which will be mainly affected by the recent selection of resources,while the original interests of users will also have some reservations,thus introducing two adjustment parameters respectively to increase the time effect of user's preference for resource selection.The algorithm has a significant improvement in the accuracy and diversity of the recommendation,indicating that using this method for recommendation can effectively improve the recommendation performance.(2)Researched the label feature of users and the importance of individual users,a recommendation algorithm based on multi-label clustering and core users is proposed.First of all,this paper taking into account the importance of individual users,clustering user labels using density peak clustering algorithm.Then,the potential relationships between users and labels are studied and we proposed a method based on multi-label clustering to determine core user,and define the concept of correlation between user and label cluster,user location weight,then based on this,a recommendation algorithm based on multi-label clustering and core users is proposed.According to experiments demonstrate the effectiveness of the proposed algorithm,we have a more significant upgrade in the recommendation accuracy and diversity.(3)Design and implement a prototype system of news recommendation based on user feature.Using the above two algorithms as the theoretical basis of the news recommendation prototype system,first analyze and design the corresponding functional modules of the system,and then implement a news recommendation prototype system based on user feature.The system can exploit the user's interest according to user feature,thereby made it easy for users to find news information that they are interested in or potentially need.