The Research For User Recommendation Algorithm In The Sina Microblogging Social Network

With the explosive growth of users on the microblogging platform, its users createmuch information which is growing exponentially, so that they can not get the informationthey want. The utilization of information is reducing and the problem of informationoverload is increasing. The current search engines and other technology can only meet thepart of demand of the people. Without personalized consideration, it still can not solve thisproblem effectively. Users recommendation as a means of information filtering is verypromising to solve this problem. Thus, how to develop efficient, scalable, very preciseuser recommendation algorithm will be a huge challenge.In this paper, two user recommendation algorithms are proposed, which are based onthe characteristics of the currently popular microblogging platform. One recommendationalgorithm is based on the preference degree in the domain of celebrity, the other is basedon the ability of information dissemination in the user community. The first algorithmconsider user recommendation as into a classification problem based on based linkprediction. It fextracts a range of features around the target users who are recommended toaccording to the donain the celebrity user belongs to, which build an n-dimensional vector,and gets the set of celebrity user recommended for the target user by the classificationfilter the celebrity set. The second algorithm discoveries the user community on themicroblogging platform and mining the middle users of messages who have strong abilityto communicate messages according to the user’s own characteristics as well as his abilityto communicate in the community，with analyzing the flow of messages in the community.Finally it will select the optimal subset of the middle users from the whole middle users bythe relationship with a specific user community, and then recommend them to the targetuser. Meanwhile, in order to solve the problem of massive data processing, the tworecommendation algorithms also are parallel to achieve based on the Map-Reduce.Implementation and testing verify the feasibility and effectiveness of the tworecommendation algorithms about data sets on the microblogging platform. Generalassessment method based on the recommendation algorithm, the effects of tworecommendation algorithms have improved, compared with other commonly usedrecommendation algorithm, and the performance is significantly higher than thestand-alone environment, which is based on Map-Reduce.