Information Diffusion Modeling And Influence Nodes Research On Microblog

Microblogs are beloved by users since its advent because of its timeliness and convenience,and it has became users’ favorite among the network applications. The enormous amount ofusers make microblogs became an important platform for the spread of public opinions. Inaddition, microblogs create a mount of huge influence users. However, microblogs bring not onlyconvenience for communicating, but also many new problems in public opinion controlling andnational security. Therefore, it will be a great significance to study information disseminationprocess and influential nodes in microblogs. There are still some disadvantages in recent studies:1) Information diffusion model can not reflect the result of dissemination of informationcomprehensively.2) Excavating influential users using quantity of friends can not embody thecharacteristics in microblogs.3) Existing research is still lack the establishment of influencemaximization algorithm for microblog network. To resolve the problems above, this dissertationrelies on the National “863” Program, carrying out researching information diffusion and nodeinfluence in microblog network. The main achievements of this dissertation are as follows:1. An information diffusion model based on behavior prediction is built. Firstly, we analyzethe factors that decide user’s retweeting behavior, and extract a series of numerical characteristics.Then, using classification method in machine learning, a user retweeting behavior predictionmodel has been established. On the basis of the prediction model, the rules of informationdiffusion which describe the behavior of groups of users in the microblogs have been designed.Lastly, a threshold based propagation model has been established. Simulation result illustrate thatthis model can forecast the trend of information diffusion well. At the same time, we also foundthere are some influence users that play a key role in information diffusion.2. A new users’ influence rank algorithm is proposed. On the basis of the definition ofinfluence, we analyze three factors that determine the users’ influence. Then, an influence rateconcept which describes users’ influence associated is proposed. On the basis of the influencerate, we established a weighted influence model. Finally, a new microblog users’ influence rankalgorithm MIR-IM (Microblog Influence Rank based on Influence Model) is proposed.Simulation result illustrate that comparing to the PageRank algorithm, the MIR-IM is moreeffective in excavating influence user.3. An influence maximization algorithm for microblog network is proposed. We analyze thetwo main issues for influence maximization in microblog, including:1) to build an influencepropagation model that meet the characteristics of microblogging;2) to design an efficient Top-Knodes selection algorithm. An extended LTM by leading into the concept of influence rate is built,and then a heuristic algorithm framework HGAE is designed. The Simulation result shows thatHGAE outperforms the traditional algorithm.