Research On Information Diffusion Model For Micro-blog Opinion Leader Mining

As a new carrier of public opinion and a novel platform for information diffusion,the micro-blog plays a more and more important role in the spreading of the online public opinion.Because of its new features such as real-time,autonomy,and interactivity,the micro-blog can contribute to the rapid spread of information and influence.Therefore,based on the information diffusion network and characteristics of the micro-blog,the information diffusion model suitable for micro-blog network is studied and the method of opinion leader mining is proposed based on the information diffusion model.Firstly,due to the massive number of micro-blog users and the real-time information diffusion,it is difficult for the traditional model to characterize the process of the influence diffusion in the social network constructed by micro-blog.In this thesis,two information diffusion models,namely: Extended Independent Cascade Model(EIC),Multi Attribute Dynamic Fusion(MADF)are proposed to model the influence diffusion process in micro-blog.The EIC model based on the traditional Independent Cascade model,when describing the influence of communication between users,considers the characteristics of the network structure,individual attributes and behavior features.Besides,the influence strength between users are expressed as a value combined by above multiple features using AHP.The MADF model makes use of decision tree algorithm to calculate the information entropy of each attribute to objectively measure the influence strength of them.And the information fusion theory is adopted to caculate the influence between users,which fuses the available influence factors in micro-blog,such as the number of users' fans and micro-blog.Secondly,a weighted diffusion network is established based on individual attributes and individual interactions in the process of information diffusion.And based on the above information diffusion model,an improved opinion leader mining algorithm is proposed.In the process of dynamic information diffusion,we can quantitativelyanalysize the user's influence ability to mine the most influential node set under the specific topic of micro-blog.Finally,experimental results of real data collected from Sina micro-blog show that the information diffusion model suitable in the micro-blog network proposed in this paper can effectively reflect the process of influence spreading among micro-blog users.At the same time,the opinion leader mining algorithm not only can effectively dig out the opinion leaders,but also can have better performance.