Research On Rumor Detection Model Based On Text Multi-feature Fusion

With the rapid development of the times,social platforms have become important channels for people to publish and obtain information.Social platforms have brought people a lot of convenience,but social platforms have also greatly increased the number of rumors and the speed of dissemination.Most netizens are unable to distinguish fake news from this information,which has caused considerable harm to the society.In the face of the rapid spread of a large number of rumors,how to automatically identify rumors and minimize the harm they cause to society is particularly important.For this reason,in order to be able to automatically identify rumor information,this article designs a rumor recognition model based on sentiment feature analysis and a rumor recognition model based on time sequence feature analysis from two perspectives of sentiment feature analysis and time series feature analysis,and designs on this basis A rumor recognition model with multiple feature fusion methods is developed,and experimental analysis is made for each model.The main work of this paper is as follows:(1)In the BERT model,the loss of sentiment features of the masking language model is large and the sentiment words of the text are less.A rumor recognition model based on BERT(LabelBERT)is proposed: the model first uses the sentiment dictionary to add weight to sentiment words to improve BERT's prediction.Training tasks to improve the model's ability to extract implicit sentiment features.In addition,a new text sentiment polarity prediction task is designed to strengthen the model's ability to acquire sentiment features for the text context,and at the same time combine BI-LSTM to identify false information.(2)Aiming at the problem of ignoring the potential information and partial user information of the text and the timing relationship of microblogs in the same event in the process of sentiment feature extraction,a rumor recognition model based on two-layer Attention-BIlstm Time series features is proposed.(Time series feature analysis Rumor Recognition model,TRRM): First,input the text into the first layer of Attention-BIlstm to get the text feature vector,use feature engineering to extract the text user features and event features,get the time series feature vector,and fuse the two to get the time period feature vector.Then send this time period feature vector to the next layer of Attention-BIlstm,learn the hidden layer feature relationship between each time period,and then realize the rumor recognition of the text data,and finally output the rumor recognition label.(3)In addition,the advantages of the LabelBERT and TRRM models proposed in the paper are fused,and three feature fusion methods are used.The first is simple feature fusion,which connects the feature vectors;the second introduces gating Mechanism for feature fusion;the third is to introduce a feature fusion method based on the channel attention mechanism,so that the model can assign weights to multiple features,pay attention to more important features,and propose a multi-feature fusion rumor recognition model(Rumor Recognition multi-feature Fusion Model,RRFM).Finally,a comprehensive experiment was carried out on the commonly used rumor detection data set.The results show that,compared with other comparison models,the multi-feature fusion rumor detection model combined with the channel attention mechanism has better results.