Research On Offline Algorithm For Mining The Social Relationship Among Microblog Users

With the rapid development of Internet and the mobile communication technologies, more and more people were attracted to various kinds of social network platforms. Complex social network graph is composed of a large number of users and complex relationships and links among these users. It becomes a hot spot in the research field how to mining valuable information from such complex network. Mining microblog users’social relationships has great significance. It can estimate the strength of potential relationships among users and find out the target user’ circle of potential friends. Then it can facilitate the implementation of user tracking, link prediction and friend recommendation. So based on the related research, this thesis proposed a new algorithm model for mining microblog users’social relationships. This new algorithm model can estimate the relationship strength among microblog users comprehensively and effectively, and then it can find out the users’circle of friends.In this paper, the major researches are in the following areas:Firstly, this thesis described the research status of mining social relationships, including two parts:evaluation indexs of social relationships and evaluation model. And this paper present that the existing algorithms had the following disadvantages:the quantity of the evaluation indexs was not enough, microblog network characteristics were not completely considered when calculating the relationship strength among microblog users, and user behavior characteristics were not considered when fusing data from various evaluation indexs.Then, this thesis proposed an offline self-adaptive algorithm model for mining social relationship among microblog users based on microblog network data with multi-dimensional user data fusion. This paper proposed five-dimensional user relationship evaluation algorithms using Virtual Microblog Graph, POI data, Virtual User Graph, User-Microblog Bigraph and background information. And this paper dealt with these five-dimensional relationship strength values with the decision algorithm model based on maximum likelihood (DAMML). Through analyzing the decision results, this thesis can find out the potential social relationships among microblog users.The five-dimensional algorithms are as follows:● The evaluation algorithm of relationship strength between users based on interests of users and virtual microblog graph (IUVMG). This algorithm added users’comments and praised information to the user interest feature vectors. And it raised the evaluation accuracy of user interest relationships.● The evaluation algorithm of active location similarities between users based on administrative region and user check-in information (ARUCI). This algorithm included subalgorithm for recognizing user active location and subalgorithm of similarities of user location based on encounter time and visit frequencies of friends. It solved the problem that some users didn’t have enough check-in information in microblog network.● The evaluation algorithm of relationship strength between users based on out-in degree and mutual friends (ODMF). This algorithm introduced out-in degree information and virtual users. And this algorithm expanded it to second degree friends.● The evaluation algorithm of interaction closeness between users based on tendency of user interbehavior (TUIB). It solved the problem of the interaction inequality between microblog users.● The evaluation algorithm of relationship strength between users based on user background information (UBI). This algorithm introduced principal component analysis in order to reduce the correlation of user background attributes and improved its accuracy.Finally, this thesis developed a microblog data crawler and acquired a large number of data from Sina Weibo. This thesis conducted experiments based on the algorithm model proposed by this thesis and some other classical algorithms. After comparing the experiment results of this paper’s algorithm model with other experiment results, it was found that the algorithm model proposed by this thesis had better performance than other classical algorithms. Besides, according to the algorithm model proposed by this paper, this thesis designed and developed an offline mining system for mining social relationship among microblog users based on B/S architecture.