Individual Opinion Prediction On Agent-based Model In Complex Social Networks

With the development and popularity of mobile devices and social network,more and more people are becoming accustomed to use social platforms to communicate with each other.However,a new group named "keyboard man" is emerging because of the anonymity of social network.And extreme events occurred frequently with the reason of herd mentality,but the fast diffusion of information in social network makes the department can't take preventive measures immediately.Aiming at these problems given above,the prediction of individual opinion related to a topic in social network is sorely in need.As Shakespeare said,there are a thousand Hamlets in a thousand people's eye.Different individualsalso hold different opinions on the same topic.Therefore,due to better understand the various influencing factors of self-opinion formation process,we consider the concept of Agent in complex adaptive theory,and proposed an Agent-Based model and Multi-Agent based opinion dynamic model.The opinion dynamic of the topic community would be saw in a micro perspective,and we can predict the future tendency of the individual opinion on the same topic.In order to achieve a better forecasting effect,we first mining several topic communities from the real dataset,and established a Multi-Agent environment based on one topic.On this basis,we also mining the implied real opinion of all the text in the topic.Then established the dynamic model of the topic community to simulate the relationship and interaction between agents.After that,established the Agent-based model and designed Agent inner structure.An opinion of one agent would be generated through the parts of sensor,controller and effector.So Agent have autonomy,adaptability,socialization and interaction characters in the Multi-Agent environment.And for each agent,three characters of individual self-confidence,neighborhood dependency and community convergence have been considered as the influential factors of forming process of the opinion.In order to realize communication and interaction between agents,we also established the Multi-Agent opinion interaction model,and designed an algorithm of generating the opinion dynamic rules.In the end,we designed and implemented experiments based on Weibo dataset to verify the validity of Agent-based model and Multi-Agent based opinion dynamic model.The results are as following:(a)From the microcosmic perspective of view,this two model can be better predict the trend of individual opinion accurately,and there is only a marginal difference between the predictive result and the actual result on the active individuals in the topic community.(b)From the macroscopic point of view,it is possible to better predict the changing trend of the topic community based on our models.