Research On Technique For Mention Target Recommendation In Twitter Based On Multi-Factors

Looking back to the development of social networks,such as BBS,MSN,QQ,WhatsApp,Facebook,Renren,Twitter and Sina Weibo,products and forms of Social network are getting more various,Social network ecosystem is becoming more complete,and provide all-day social network's life without interruption.Social network has been from the initial "incremental entertainment" gradually become "constant life",that led to the explosive growth of social network information.Overload problem also arising,users of information screening criteria Higher and higher.How can meaningful information is delivered to others on social networks faster and be recognized?Today,Twitter has become one of the most important platform to share information,Mention is a special feature in Twitter.Tweets can be used in tweets while Twitter users are in the middle of tweets,but when a tweet is sent,the user is notified and can push tweets and / or other users,so that the user can spread out.Obviously,finding the right user to do it can increase the spread of tweets and the impact of users.How to find the right user is the main content of this paper.After reviewing the research situation and related theories on information recommendation,this paper presents a multi-factor user recommendation model for Twitter.The main contents are as follows:(1)Analyzing and validating influencing the influence of factorsAfter understanding and analyzing the Twitter-like microblogging system and its user recommendations,four factors were identified that were most likely to influence user behavior: interest similarity,user vulnerability,users' online,and user GEO Location similarity,and verifying the significance of the various factors on mention user recommendation and the actual impact of the situation.(2)Multi-factors blended recommendation model based on learning to rankAfter verifying the impact of the four factors on the user,the eigenvalues of the factors are constructed according to the experimental data,and then by the sorting function,the first three factors are determined by the fact that the factor "user on-line situation" can not be quantitatively measured.The weight of each feature,the initial user fusion model that combines the three factors of interest similarity,user fragility and geographic similarity is formed,and the validity of the model is verified by experiments.Putting a pending tweet and a user in the input model,you can calculate its recommendation score.(3)Recommendation optimization model based on assignment problem modelBecause users' online status can't be measured.In order to tweet @ online users is more difficult.Here we divide the day into 24 time windows by hour.We estimate the online probability of each window by the user's history in each window,and then sum them with the scores in the multi-factor fusion model.Utility value.After forming the utility matrix for all candidate users,the problem is solved based on the assignment model,and the result is the user's scheme.The experimental results show that the proposed model can effectively improve the diffusion efficiency.In this paper,the impact factors of influencing user selection are studied in the context of user recommendation in Twitter,and a multi-factor fusion recommendation model based on ranking learning is proposed.By adding user online factors to the model,we construct a recommendation optimization model based on assignment model.The effectiveness of the proposed model is verified by compared experiments.