Research On Microblog Rumor Detection Model Based On Difference Analysis

With the advent of the information age,more and more people share and obtain information on social media.Rumors are often mixed in this information,which greatly affect people's daily life,and some even cause social panic and affect the social stability.Considering that Sina Weibo is currently the most active social media in China,so we choose Sina Weibo as our research object.Traditional rumor detection methods usually use traditional machine learning models to extract event features by designing artificial features,which ignore the semantic features of the text information.Compared with traditional machine learning models,deep learning models can automatically extract features from a large amount of text information,which has the advantages of low labor cost,simple training,strong generalization ability,and good detection effect.Therefore,we conduct research on deep learning models.In view of the fact that most of the current rumor detection models based on deep learning consider the different types of text information of the event as a whole,which only focus on the characteristics of the event itself.Therefore,we analyze the differences in the text information of the events,the differences in the representational characteristics of the events,and the association of the events and the differences in this association.Our main work are as follows:(1)Aiming at the differences in the text of the events,we propose a feature extraction model of microblog rumors based on content differences The model first converts the text of the events into vector representations through pre-training method.Next,different methods are used for feature extraction on the original text information and comments text of the events,and we obtain the original characterization feature and the comments characterization feature of the events.At last,we obtain the event characterization features considering text information differences by splicing the characterization features we mentioned before.(2)Aiming at the difference in the representational characteristics of the events,based on the feature extraction model of microblog events based on text information differences,we introduce the attention mechanism and propose a microblog rumors detection model based on feature differences.This model assigns different characterization weights to each dimension of the comments characterization feature of the events through attention mechanism,and then we obtain the event characterization feature considering representational characteristics differences by combining with the original character characterization feature.At last,we use this model to complete the task of rumor detection.(3)Aiming at the association of the events and the differences in this association,we introduce the graph convolutional neural network,and propose a microblog rumor detection model based on event associations and their differences.This model constructs a microblog event association graph by considering events as nodes,the associations between events as edges,and the closeness of the association as the weight of the edge.We take the adjacency matrix of this graph as the input of this model,whose outputs are the category information of the events nodes through the graph convolutional neural network.At last,we use this model to complete the task of rumor detection.