Research On Fake Information Detection Method Oriented To Time And Structure Characteristics Of Microblog Information Dissemination

As an emerging media of the information age,social media has gradually occupied people's lives,and replaced traditional media as the mainstream of people access to information.However,the real-time dynamic characteristic of social media has not only brought convenience of the high-speed flow of information,but also makes social media becomes the hotbed for the birth and dissemination of fake information,which effects people's normal life and social order strongly.How to quickly and accurately distinguish the true and fake information from the massive information on social platforms and then to prevent the spread of fake information has become a hot topic for many researchers in recent years.In this thesis,we studied the blog data of “Sina Weibo”,the largest social platform in China,analyzed the information event propagation process of Sina Weibo from time and space dimensions,explores the changes of feature in different propagation modes.Then different modeling methods are proposed based on different propagation modes,the methods results show that the performance of fake information detection is improved.Firstly,aiming at the temporal propagation mode of information events,the LSTM(Long Short-Term Memory)is used to mine the changes of feature over time.At the same time,in order to make full use of the text content of the information event source,the CNN(Convolutional Neural Network)is used to mine its high-order semantic information.Based on this,a multi-module neural network detection algorithm is proposed,which combines LSTM and CNN,makes full use of the good ability of LSTM to deal with timeseries problems and the advantages of CNN in natural language processing tasks.Secondly,aiming at the information event topology propagation mode,this thesis firstly studies the effect of information event propagation tree structure in fake information detection problem,and then improves the existing algorithm,optimizes the definition of nodes and edges of propagation tree,and strictly restricts the simplification process of the propagation tree,which makes it possible that the emotional diversity of the tree is preserved while simplifying the propagation tree.In addition,in view of the fact that most of the current research work only uses a single classification model,the effectiveness of the ensemble learning algorithm for fake information detection is studied under the given information event topology propagation mode.