Opinion Spreading Models On Complex Networks

From the view point of idea that system functions are determined by their structures, this thesis empirically and theoretically study the opinion spreading and guide strategy on complex networks. With the rapid development of information technology and Internet, as a publication and dissemination of the carrier, network is playing more and more significant role in opinion spreading, including the real and groundless informations. Development of information technology has brought convenience to people's lives, and provides a convenient way for the spreading of rumors and information related to public safety. As an effective tool to descript the real-world systems, complex networks have been used quantitatively. If the "personal" and "relationship" in the online social network system are abstracted as nodes and edges, respectively, complex networks can be used for modeling and analysis the opinion spreading process of Web online social networks.Firstly, an online social network data which has 3.54 million users are statistically analyzed, where the users and relationships are defined as nodes and edges respectively. The statistical results show that this system star-type structure which is totally different from the small-world and scale-free networks, which has joint topology of the core node connected to each other and their in-degree distribution approximately satisfies the Zipfs law and the sparsity is increase linearly with the network size. The classical opinion spreading model with independent update rule is implemented on constructed network and the simulation results show that the change value of all users'opinions will be quick. This results show that opinion spreading speed on real-world system more quickly than the one on small-world and scale-free networks. Furthermore, the effects of the local and global important nodes on the opinion spreading are investigated. The local important nodes could be introduced by their degree. The simulation results indicate that controlling the local important nodes'opinions could not speed up or slow down the opinion evolution process, but the system would evolve more smoothly. In addition, the effects of the global important nodes are considered, the simulation results show that controlling this kind of nodes could slown down the spreading speed greatly. Since much time must be spent to comupute the node betweenness, which has hindered its application.Secondly, two improved models are studied theoretically and numerically. In the first model, the accepted probability is introduced based on the phenomena that each user would accept his his friends'opinion with probability instead of fully believe or accept. Theoretical analysis and simulation results on grid network show that the proportion of various points of view will always be maintained at a stable level. On a specific probability, different views can coexist on the random networks, otherwise, all the people will hold the same views. In another model, we assume that, beside the persons whose opinions are +1 or -1, there exists the third kind of person, which can be affected by the neighbors and this kind of person would communicate with his neighbors and accept a friend's point of view with a probability. Analytical analysis and simulation results show that the proportion of opinions with +1 or -1 only correlated with the initial proportion of the persons whose opinions are +1 or -1, but has nothing relationship with the exchange probability. The existence of particular network with specific parameters can reach equilibrium.Thirdly, the opinion spreading model based on the link prediction mechanism is proposed. Based on the mass diffusion and heat conduction processes, a hybrid algorithm is proposed which has higher accuracy and diversity. The improved algorithm is implemented on Livejournal data to study the opinion spreading. Since the livejournal data evolves dynamically and is only a sample of the real data, the opinion spreading model implemented on the network with predicted links would be more practical. Empirical results found that the real opinion would spreading more quickly on the network with predicted links, which indicates that communication between the role of the core nodes would enhance the spreading speed and range greatly. In addition, the simulation results indicate that decreasing the influence of the core nodes could increase the spreading speed correspondingly.