Analysis Of User Influence Based On Topics On Twitter

Micro-blogging service has become the representative application of social networking due to its opening,easy operation,wide interactivity and a wide range of information,and has increasing impacts on people's daily life.Micro-blogging information,user information,topic information and interactive information are springing up rapidly,with users' interpersonal communication,opinion expression,information acquisition,information sharing and other behaviors on the Internet,users gradually obtain their own virtual society status in the social network.In the social network,users from different positions have different effects on the distribution of information,as well as different kinds of information.Therefore,research on the influence of users in micro-blogging,especially the topic-oriented influence of users,has important research significance and practical value.Micro-blogging contains massive data,but a great amount of statistical noise,the complexity of user relationships,and the rapid evolution of information diffusion varies over time and other characteristics have brought great challenges to this study.The main work of this paper is to extract the user attributes and relational attributes from the data with statistical noise,analyze the impacts of users in various fields of topics,and then find influential users.Meanwhile,in order to achieve the purpose of predicting the diffusion ability of user information,the author uses data mining and machine learning methods to analyze and establish the prediction model of the diffusion of users' influence in micro-blogging.Following provides a brief description of the main contents and innovations of this paper.(1)This paper defines the meaning of user influence in micro-blogging and how to measure the distribution of the influence of users.(2)Moreover,the author extracts user static attributes,behavioral attributes and relational attributes from the collected micro-blogging user data,and applies LDA model to model the topic space and calculate the distribution of users' blog post in each topic based on the given number of topics.This author exploits the topic-based UAMR InfluenceRank algorithm to analyze the influence of users in various topics and chose a set of real Twitter data.The study verifies that the UAMR-InfluenceRank algorithm is superior to PageRank,TwitterRank and other classic data mining algorithm when calculating the influence of Top-N users based on several evaluation index such as accuracy,recall and F-value.(3)This paper uses data mining method to conduct data preprocess and feature extraction from the Twitter data set,and obtains the characters which can reflect user influence and diffusion from users and their blogs.Consider the xgboost algorithm as the base model,which is applied under the condition of multiple feature groups,multiple sets of parameters and data set time-sharing.Based on the idea of Stacking,the author proposes a Multi-View Multi-Parameter Time-Division Stacking Ensemble algorithm and model to predict the diffusion of users' influence.The Twitter data set not only verifies the feasibility and effectiveness of the proposed methods,but also proves the method to learn features of micro-blogging and predict the diffusion effect of future message updates published by users is effective according to user's historical information.In conclusion,based on the massive micro-blogging data,this paper studies the topic-oriented influence of users from multi-user attributes and multi-relational attributes,and adopts data mining methods to model and predict the distribution of influence of users in micro-blogging.It has important practical value to the analysis of public sentiment and key users profiling in network.