Research On Mining Structure And Behaviors In Mobile Social Networks

With the rapid development of mobile internet, people’s social approaches havechanged from traditional social website to mobile area. Thus the mobile social networkswith timeliness, mobility and customizability came out. As the two major applicationsof current mobile social networks, message and microblog cover the most popular usergroups and have the good representation. Therefore, the research on the mobile socialnetworks constituted with message and microblog will help to find the characters of thevirtual society and its representation-the real society. The characters will not only helpto consult the research of different relation attributes in the same network, but alsouseful to the research of the same identification in different networks.On the other side, the rapid development of mobile social networks increases theinformation explosion problem which is already exit. Since the explosion of vastquantities of information, opportunities for the liberalization development ofinformation have been brought out. While, a large number of garbage information alsocomes out at the same time. If the existence of the garbage information is indulged, theusers will suffer larger loss and the development of mobile internet industry will beseriously hindered.Based on the above reasons, this thesis will firstly study the structural andbehavioral characteristics of mobile social network. Then using the features to predictthe multi-relationship within networks and identify users from different networks.Finally, a spam filtering framework is built for mobile social network by combining thefeatures and information processing technologies. The main results of this thesis are asfollows:1) The structural and behavioral characteristics of mobile social network especiallyon short message and microblog is researched. The differences between the mobilesocial network and traditional social network, as well as domestic network and foreignnetwork are analyzed from a variety of perspectives.2) The kinship structure and behavior characteristics is studied in message network,and some salient features are further extracted. Then, a kinship model is built on those features.. Based on the model, the existence prediction (whether there is) and the typeprediction (what kind of the relationship) of kinship between two users with interactiverelationship is performed.3) An identity de-anonymization method in multilayer social networks is putforward according to the situation that one user may exists in multiple mobile socialnetworks. By building lowest energy model according on Ising model, the user energyin the multi-layer networks is represented. The corresponding relations prediction andreconstruction of identification of user in different layer networks according to thismodel is implemented.4) Since the concept drift problem existing in traditional spam filtering process,namely the underlying concepts in message stream will change along with time andenvironment. The original filter system cannot automatically adapt to these changes.Furthermore, filtration velocity and filtering accuracy cannot be given equal attention inthe current conditions of the massive data. In order to resolve this problem, theself-feedback spam filtering framework and flow filtering algorithm which can realizethe real-time and intelligent mobile social network are presented.