Research On Pivotal Issues Of Users' Emotional Behaviors In Social Network

With the rapid development and maturation of Web2.0 and mobile communication technology, social networks, such as Twitter, Facebook,Sina-microblog, have gradually integrated into people's daily life. Users of various social levels are gathering together in social networks according to their particular demands. Social networks make time difference, space difference, etc, no longer be communication barriers, and extend the social relationship on the virtual network. With the vigorous development of social networks, dissemination of information becomes more pervasive and widespread. Massive users in social networks could easily browse news,discuss hot topics, interact with friends or strangers, and they could express their various emotions at any time through posting, commenting,forwarding, etc. Communications in the internet are always considered as fictitious. However with the spread of emotional information in the network, different users' emotional interactions arised, and users' behaviors in the real world even could be affected. The research of human emotional behaviors has attracted many scholars' interest that from sociology,psychology, economics, computer science and other disciplines. However,due to the complexity of human emotions, the researchers are facing various challenges. Massive data of user behaviors in social network is recorded in real time, this give us an unprecedented opportunity to research human emotional behaviors. The research of users' emotional behaviors in social network has great significance and wide application foreground.This dissertation is offered as a preliminary probe into the research on users' emotional behaviors in social networks, and the complex network theory and data mining methods are employed in the research process. The main research contents include: 'analysison of graded emotional behaviors and its modeling and simulation'; 'users clustering based on multivariate time series of emotional behaviors'; 'a model for finding emotional influencers'; 'users emotional community detection'. The main work and innovation of this dissertation are as follows:(1) Based on the massive Sina-microblog data, an emotional stage posting model is proposed and the validity of the model is verified by simulation. Detailed process is as follows: firstly, microblogs are graded by emotional stage, and the users' emotional behaviors are analyzed. The analysion found that posting amount of users obeys power-law distribution at a certain emotional stage,and exponent increases as the emotional stage tends to calm. Most of posts are peaceful, when the intense emotions need to be expressed, users' participation rates will decrease; secondly, the model is proposed and the impact of the around emotional environmental factors of the poster is considered, and also the randomness changes of poster's own emotion; finally the power-law distribution and the variation tendency of power exponent are verified by model simulation.(2) Based on the analysison of users' multivariate emotional behaviors,a novel similarity measure of multivariate emotional time series (METS) is proposed, and users in social network are clustered and analyzed. The detailed process is as follows: firstly, microblogs' multivariate emotion vectors of every user are extracted and METS is constructed by these vectors; secondly, a novel similarity measure is proposed which is a combination of PCA similarity and distance similarity. The combined similarity measure considers both emotional fluctuation and intensity;finally the combined similarity is used in k-means method, different user emotion groups and their emotional behavior characteristics are found in clustering experiment.(3) Based on the heterogeneous characteristics of social network and emotional interaction between network users, a model (EmotionRank) for finding emotional influencers in microblog is proposed. Detailed process is as follows: firstly, a heterogeneous microblog network that contains two types of nodes (user, microblog) and three types of relations (forwarding,following, posting) is constructed. Emotional homophily is verified in this network, which confirms the existence of emotional influence; secondly,the heterogeneous network is transformed to be a homogeneous network that contains only users, and the random walk model is employed to find emotional influencers in this network; finally, experimental results based on microblog data effectively illustrate the utility and superiority of EmotionRank.(4) Based on the emotional homophily of users in social networks, it could confirm that users would like to gather together to form communities according to the emotional similarity. Using the social network's structure,this work proposes to construct an emotional network model, which employ users' and microblogs' emotional similarity as edge weight. Both CNM and BGLL algorithms are employed to detect emotional communities in this network. In order to verify the superiority of emotional network when detecting emotional communities, it is compared with 4 networks (1 unweighted network and 3 weighted neworks which based on 3 different similarities of nodes in network). The 5 networks have the same structure and different edge weight. Comparison results illustrate that both users'emotional behaviors and the microblogs' emotion are more similar in identical communities of emotional network.