Early Microblog Rumor Detection Based On Sentiment Analysis And Transformer Model

This article is based on the Sina Weibo platform,with the goal of realizing the early detection of rumors,in-depth mining of the semantic information of the content of the Weibo text,and emphasizing the emotional orientation of users in Weibo comments,building a Sina Weibo rumor detection model,Has improved the efficiency and accuracy of rumor recognition,and mainly carried out the following three tasks:The first work is to draw lessons from the idea of ??pre-training,by pre-training some model parameters,improving the training speed of the model,while solving the problem of word vector representation in the case of polysemous words,optimizing the expressive ability of deep semantic information,and further Improving the timeliness of rumor detection has laid a good foundation for the early detection of rumors on Weibo.The second task is to change the six-layer structure of the Encoder model in Transformer to two layers,which improves the training speed of the model,and uses the two-layer Transformer encoder to extract the semantic features of the content of the Weibo body.The attention mechanism can well capture long-distance sequence features and improve the ability to understand text semantics.The third work is to use the Bi GRU+Attention network to extract the emotional features of Weibo comments.Then,the semantic feature vectors of the Weibo text and the emotional feature vectors of the Weibo comments are spliced ??and merged to further enrich the input features of the neural network,and then output The classification result of the microblog event,and then realize the microblog rumor detection.Experimental results show that the two rumors detection methods proposed in this article: XLNet+T and XLNet+TAB are both true and effective on the Sina Weibo data set.Among them,the accuracy rate of the XLNet+TAB model can be as high as 95.9%,which can be used in early rumor detection tasks.Above,the two methods proposed in this article can effectively solve the early rumor detection problem in terms of the rumor recognition performance under different deadlines.