Research Of Microblog Friends Recommendation Algorithm Based On Trust And User Behavior

With the rapid development of the Internet,people's daily life has changed greatly.Such as the emergence of Internet application has simplified the way that people obtain information.More significantly,the emergence of social applications makes the way that information transmission diversified.Admittedly,Social networks have gradually become a new platform for information acquisition and communication.Weibo,as the most popular social media nowadays,attracts a lot of users to share information and communicate with friends.As the number of the users and the amount of information grows so fast that it becomes more and more difficult for users to find what they are interested in.Therefore,finding potential friends for users becomes the main personalized service of microblog.Firstly,the service finds similar users through the user's personal background information,relationship network maps,historical behavior information and so on.Then it recommends the user's friends to develop their own social relationships,thereby enhancing the user's activity and adhesiveness on the platform.So,this thesis proposes two kinds of friend recommendation algorithms based on the user's different purposes for the current status of the research of Weibo friend recommendation,the details as follows:1.The existing researches of microblog friend recommendation do not distinguish the user's attention on the purpose of following friends.And most of researches use the same recommendation algorithm for the user or the community.No doubt,the final recommendation result lacks accuracy and diversification.Therefore,according to the preference for attention or shareing their content that users are interested in,the thesis proposes a measurement method of user similarity-degree and confidence-degree by the use of reducing the influence of the well-known user recommendation,combining with the logistic regression training model,and microblog friend recommendation algorithm,is established based on a trust-based user relationship.Then,the breadth of recommendations and the accuracy of recommendations are guaranteed,on the basis of satisfying the user's attention.2.Collaborative Filtering recommendation algorithm mainly uses scoring data to construct scoring matrix,so as to carry out users rating prediction,and then generate recommendation results according to the prediction score.However,there is no rating data inthe microblog.So it is difficult to use this algorithm for recommendation.Concerning on this problem,this thesis proposes a method of using user interactive data,such as retweet,like,and comment,to convert to user rating,and then constructs a scoring matrix and combines SVD++ model to mitigate matrix sparseness problem to obtain the interuser prediction score.3.Considering users' preferences for timely understanding of hot events,a friend recommendation algorithm based on user behavior is proposed,which adds user influence based on user rating.Because the users who publish hot events often have high influence,it should increase the proportion of users who publish hot events in the recommended results,under the premise of ensuring the accuracy of recommendation.Therefore,the final results verify that,in microblog friend recommendation,using collaborative filtering algorithm can effectively realize recommendation behavior without scoring data.It is also proved that the two algorithms presented in this thesis have advantages over the single factor recommendation and the recommendation algorithm without considering the users' purpose in the real data of microblog,which effectively improve the accuracy of the friend recommendation.