Research Of Behavior Evolution Analysis And Its Application For Internet Public Opinions

With the rapid speed development and popularization of Internet in the global scope,network media has already been recognized as “the fourth media” after newspaper, broadcast andTV. Network media has become a new platform for exchanging ideas, clarifying points andpublishing comments, due to its open, quick and interactive features. Under the action ofnetwork media, Internet public opinions have faster propagation velocity, broader diffusion areaand bigger influence strength than traditional social public opinions. So it plays an improvingrole in expression and propagation of public opinions. As the virtuality of network media andlack of effective supervision, network media becomes the space of emotion abreaction, thus maymislead people. It will make great impact on society unless taking effective measures to controland lead it. The research of behavior evolution analysis for Internet public opinions can providetechnical support to the relevant departments of the state understanding the public opinions,discovering and controlling Internet public opinions in time. The paper mainly researchesbehavior evolution analysis and its application, and the contributions are listed in the followingthree aspects:(1) Behavior evolution characteristics of Internet public opinions are researched, andbehavior evolution patterns are constructed. The paper uses the time series which describe thenetizen’s posting process to represent the behavior evolution process of Internet public opinions.By combining the methods of analyzing time series, the paper constructs six behavior evolutionpatterns, including distributional pattern, stationary pattern, correlative pattern, self-similarpattern, periodic pattern and trending pattern, and presents corresponding methods to realizethese patterns. The experiments show that the patterns can express the behavior evolutionprocess of Internet public opinions effectively, and provide theoretical basis for behaviorevolution modeling and trend forecasting.(2) The dynamic and multi-componential characteristics of behavior evolution processconsidered, the paper presents an adaptive behavior evolution modeling method and a behaviorevolution modeling method based on EMD respectively, to improve model adaptability andsimplify behavior evolution process in two viewpoints. The former method follows causalrelationship of evolution analyzing before modeling, and selects appropriate model fromalternative model bank to model behavior evolution process, by tracking behavior evolutionpatterns dynamically. At the same time, the latter method decomposes the behavior evolutionprocess by EMD to form trend component, period component, mutation component and randomcomponent. Subsequently, by summing up the predicted values of components, the method canget the whole evolution trend. The experiments show that the trend forecasting performance ofthe above two methods is better than that of the current methods.(3) The paper researches the application of behavior evolution analysis in pre-warning ofInternet public opinions, and proposes a pre-warning method based on cloud model. The methodimports the trend forecasting information of behavior evolution into the pre-warning process, and gives a threat assessment technology based on cloud model. The technology can get thepre-warning level after fusing the current threat state and the potential threat trend. Theexperiments show that the method has certain fault tolerance and robustness, and thatpre-warning effect of the method is better than the existing methods. The pre-warning levelobtained by the method is time-varying and the change process is consistent with the eventdevelopment process, which reflects the netizen’s custom of surfing Internet and accords with theobjective reality.