Research On Rumor Detection For Sina Weibo

In recent years,with the continuous development of information technology,Sina Weibo,as one of the most popular social networking platforms at present,its convenience and openness make the information spread more quickly and widely,and the rumors of Weibo are also rampant,seriously endangering the healthy network environment.The research on rumor detection in Weibo has been paid more and more attention by researchers.How to excavate the deep-seated effective features of rumors in Weibo and build a rumor detection model with good recognition effect has become the key research direction of researchers.Taking Weibo rumor as the research object,this paper analyzes Weibo’s user information,communication information and text information,extracts 11 shallow features and 3 deep text features,and selects the feature subset which has the best effect on Weibo rumor detection model based on the feature selection method,builds a variety of algorithm models and optimizes key parameters to realize the automatic detection of Weibo rumor.The main research contents include the following aspects:(1)analyzing the characteristics of rumors in Weibo from three characteristic dimensions:user characteristics,communication characteristics and text characteristics,extracting the key shallow features of rumors in Weibo,which are different from those of ordinary Weibo,and deeply analyzing the text features of Weibo’s original text and its comments,it is found that Weibo’s original text has a certain thematic tendency,and there are many comments with heavy emotional color in the comment area of rumors in Weibo,as well as rumors of rational netizens,so as to dig out the ideas of extracting deep text features;(2)Using natural language processing technology to extract the deep text features of Weibo’s original text and its comments.Firstly,the LDA theme model is used to extract the theme distribution characteristics of Weibo.Then,the keyword database and account collection of rumors are constructed,and the algorithm is designed to extract the Weibo’s questioning degree characteristics combined with the comments.Finally,using SnowNLP to extract emotional features of comments;(3)Select the extracted candidate feature set to find the optimal feature subset.Firstly,a single feature is analyzed,and the statistical method is used to verify that three new deep text features extracted in this paper have significant effects on rumor detection.Then,based on Wrapper recursive feature elimination algorithm,different machine learning algorithms are used to select the best feature sets,and the feature subset that has the best influence on rumor detection in Weibo is screened out,so as to prepare for the subsequent modeling.(4)Based on four algorithms: Random Forest,XGBoost,support vector machine and fully connected neural network,the models are respectively constructed,and the key parameters of each model are optimized,and the rumor detection capabilities of different models are compared and analyzed;At the same time,different feature subsets are selected as control experiments,and the classification model is constructed by using random forest algorithm to verify the effectiveness of the deep text features constructed in this paper.The experimental results show that on the basis of the traditional shallow feature set,three deep text features extracted in this paper are added in turn,and the model classification effect is improved,with an accuracy rate of 87.42%,which verifies the effectiveness of the deep text feature extraction in this paper.The experimental comparison results of the four algorithm models show that via optimizing a series of key parameters,such as activation function,loss function and batch＿size,the fully connected neural network model performs best in rumor detection,with an accuracy rate of 90.26%.Based on the above research,this paper analyzes and extracts key features from various angles,and carries out deep text feature mining on Weibo’s original text and its comments,and then builds a rumor detection model of Sina Weibo by using various classification algorithms based on the feature set,which can achieve a good automatic rumor detection effect,provide decision-making basis for artificial rumors,and greatly reduce the rumor detection cost.