Research On User Behavior Analysis And Information Evolution On

With the development of network and widespread application of Web2.0technology, Internet has become the common place where people look for information, express their ideas and communicate with others. Local discussion on Internet can be enhanced nonlinearly after fast spreading, and may cause online emergency. Therefore, public opinion on Internet has attracted much attention of many researchers. However, on Internet especially in online social networks, the most efficient mechanism of information propagation is post retweeting based on the user relation network, and information spreads faster on Internet than in actual society. Meanwhile, the features of anonymous and random Internet interactions, the heterogeneity and subjectivity of Internet users, and the nonlinear actions between the Internet and actual society, make information evolution on Internet more complex and uncertain. Under this situation, traditional opinion models can not effectively describe the microscopic interacting behavior between Internet users, and fail to explain the phenomena of information diffusion and evolution. For that reason, we study user behavior characteristics, information diffusion mechanism, opinion evolution pattern, spreading behavior prediction in online social networks and online forums, incorporating the research methods and solutions of interdiscipline. We concentrate on the dissipative behavior and activity change of Internet users, explore the influence of Internet special characteristics on the formation and phase transition of public opinion, and model the process of information propagation in online social networks. The work helps to understand the complex collective behavior of Internet users and the rule of information spreading and opinion evolution, provides a demonstration for investigating other self-organized complex systems, and makes a contribution to further research on control strategy of public opinion.The work of the dissertation is supported by the National Natural Science Foundation of China (No.60972012,61172072), Beijing Natural Science Foundation (No.4102047,4112045), and the Fundamental Research Funds for the Central Universities (No.2011YJS005). Main contributions of the dissertation are as follows:1. Empirically analyze the participation degree of Internet forum users, and model the dynamical user behavior; analyze users’topics of interest, and put forward a post participating model based on user interest. The results prove that users’ dissipative behavior makes the relation network have both scale-free and small-world properties as real networks, and possesses power-law degree distribution, large clustering coefficient and small average shortest path length. Users’active time follows two scaling power-law distribution. On the other side, the difference of users’ interests promotes the heterogeneity of post popularity, and causes the occurrence of hot posts. In addition, the popularity of posts is influenced by the number of active users, current number of other posts, and some other factors. Under the action of users’interests, the arrival of user’s posts follows non-Poisson process.2. Propose a dynamical model of information diffusion in online social networks. According to the retweeting mechanism based on user relations, we introduce the contacted state of an agent, and build a model of information propagation process in social networks. In the model, agents have two absorbing states, that is, the infected and refractory state. Mean field analysis and Monte Carlo simulations prove that, small spreading rate can make information diffuse widely, and the degree based density of infected agents increases with the degree monotonously. Increasing the connectivity of network causes more infected agents, but their density will not exceed the limit, and larger average network degree doesn’t always mean less relaxation time. The influence between information occurs only when different information originates in the same local neighborhood.3. Model the influence of Internet user subjectivity, system structure, and spreading circumstance on the process of opinion formation. In terms of the properties of social networks, we present the opinion model based on individual inclination, system dissipative structure, and users’hidden credibility, and describe the effect of these characteristics on the microscopic behavior. Simulation results show that individual inclination forms gradually during opinion evolution, and makes the initial majority opinion predominate absolutely. Meanwhile contrariant agents form strong inclinations of the minority opinion, preventing the convergence of macroscopic opinion. Moreover, Internet community is a dissipative structural system, and only external activation can drive the dissipative system to total consensus. If the system is not active enough, the disordering of the system may be enlarged, for the densities of two competing opinions tend to be the same. Both individual dissipation and internal or external activation may accelerate the opinion formation. In addition, Internet anonymity makes people only have the incomplete information of opponent’s personality, and therefore the macroscopic dynamics is accelerated with the occurrence of more large-scale opinion clusters. We also discuss the control strategy of promoting or preventing the opinion convergence.4. Research on the prediction of hot posts in online forums and prediction of user behavior in social networks. First, according to the participant concentration of posts in online forums, we give the definition of hot posts. We extract post’s features related to its hotness, including content influence, short-term influence and time influence, and fuse these features to predict whether a post will become hot. We use the training data to adapt the parameters, and use testing data to verify the performance of our model. The model does not depend on the time sequences of post data, and experiment results illustrate that using the early data of posts, our model can detect a majority of hot posts. Then, we analyze and quantify the content, network and time factors concerning user’s spreading behavior in social networks. For each agent, we fuse these features to generate a predictive model. We use historical post data to train the model, and predict whether each user will retweet a new post. The model integrates user’s interest and user relation topology, and experiment shows our model has both high recall and precision.