Research On Theme-oriented Microblogging Opinion Leaders

Sina micro-blog has attracted more and more users since public testing because of its low registration threshold and real time sharing of information.Micro-blog not only spreads all kinds of news at any time,but also allows users to express their opinions at any time and anywhere.This speeds up the speed of information dissemination and provides users with a platform for free discussion.However,based on the virtual nature of the network,users can easily ignore the legal and moral constraints when using micro-blog,and even more people make untrue statements,causing a bad impact.So far,the control methods of micro-blog public opinion remain at a relatively primitive level.Therefore,from the perspective of data analysis and mining,this paper mainly studies three aspects of sina micro-blog's public opinion control.(1)To determine the direction of the development of public opinion,that is,the change of the heat of micro-blog with time,to provide a key topic for public opinion control departments to focus on the development of the topic.In micro-blog,a topic often corresponds to an event,and the topic that the user publishs related to micro-blog is usually a discussion of a particular event.In order to grasp the trend of public opinion of the event,we first need to understand the various themes presented by users during the discussion process and the evolution of each theme over time.(2)Determine the current influence of the users,for the public opinion control department to provide key control of the population.Firstly,organizing the data of all the users who participated in the topic discussion to construct the user influence measurement model,then calculating the weight of each index scientifically by the method based on information entropy,and finally get the initial influence of all users as a result.Secondly,building the theme-user relationship table by combining with the results of thematic analysis,and dividing user groups based on the theme of relevance,and a graph model is established to calsulating iteratively the final influence of users under different topics according to the transitive relationship between users in different groups,and ranking the final influence according to the different topics.Those who ranked on the top are theme-oriented opinion leaders.(3)Identifying users who may have a larege impact on the topic discussion and providing the groups who should be foused on to the public opinion control department.On the basis of the indexes of user influence measurement model and the characteristics of user participation topic discussion,this paper extracts 14 characters form 5 aspects including user network structure,user's historical activity,the history of microblogging,user's own characteristics and user's topic participation,which can be used to construct random forest prediction model.And optimize the number of parameter eigenvectors and decision trees in the training model so that the model prediction effect is better.This paper collates the theoretical knowledge and algorithm concentration needed in the three aspects of the study.Finally,based on the development of the actual public opinion of "the 1 billion 200 million assets are frozen in Leshi",the design experiments verify the three aspects of this study.Python and its related tool are used to realize the capture of micro-blog data,the simulation experiment of the model and the visual output of the experimental results.The experiment shows that the opinion leader recognition method and the opinion leader prediction model proposed in this paper are effective for the topic oriented opinion leader mining.