The Method Of Local Social Network Construction And Focusing Personae Extraction In Microblog

With the coming of Internet era,the Internet is integrating into people's life gradually.Many netizens usually go shopping,make friends,and study on the Internet,and it has become a very important part of people's daily life.In the people's internet life,online social networks(OSNs),such as Sina Microblog,Tencent Microblog,Twitter,have been the more active place in people's online activities.People always make new friends and share information,such as text,images,videos etc.,in which they are interested on OSNs.Therefore,these published can reflect users' habits and interests to some extent.Now,mining effective information from massive social data is a great challenge in the data mining field,because the features of social data is the sort context,large quantities,and high real time.Facing with a large number of user data of social platform,it is the key to construct users' social graph and interests graph for improving the quality of social search in the social network.In order to solve above questions,the main research contents include as follows.1.This thesis proposed Active Friend Prediction algorithm(AFP),by improving Friend Link algorithm(FL),which is based on Link Prediction.In order to apply to the online social platform,like microblog servers,that contains sparse user attribute information,the users' OSNs are abstract into the digraph to analyzing users' similarity with the local-link feature of digraph.This paper proposes the concept about node activity.It can filter out the active users according to the ratio of node's out-degree to in-degree.On the other hand,this paper works out nodes' activity score combining with the nodes' link structure similarity of user's social network.Furthermore,extracting the active indirect neighbors which have potential relationship with user according to active score,and utilizing these nodes to construct user's highly active Local Social Network(LSN),namely Social Graph.2.This thesis proposed Implicit and Explicit Focusing Personae Extraction algorithm,named FPE.Microblog is a kind of social platform which is support by short text.The microblog posts contain personae entities that are focused by users,but these texts are always full of many noises.So this paper extracts personae entities from microblog posts which published by user and other users in which his/her social graph.Then our algorithm calculates the attention rate for personae entities according to activity values of users of target user's social graph and the count of microblog post which contain relevant personae entities.And our method regard the personae entities which have higher attention rate as focusing personae to construct user's Interest Graph.Furthermore,FPE can also extract the focusing personae of the local social network.Finally,this thesis compares our method with other link-based similarity measures in precision,recall and F-score,and respectively extracts user's focusing personae from social graphs which based on different link prediction measures.Experiments show that the improved node score calculating method has higher precision,recall,F-measure than other similarity measures.FPE can effectively extract user's focusing personae,and its precision rest with the precision of user's social graph.