Microblog Rumor Detection Research Based On LSTM Sentiment Analysis Model

Weibo is an Internet social platform with high openness and freedom.Everyone can publish and receive information on the Weibo platform.Because of the huge amount of information and the extremely low access mechanism of Weibo,the exchange of information is greatly facilitated.At the same time,it is also filled with a lot of rumors.The negative effects of producing and disseminating rumors have greatly affected the harmony and stability of the country and society and personal life.Therefore,how to automatically and effectively identify microblog rumors has always been a research hot spot in related fields.Traditional Weibo rumor identification research is mainly regarded as a dual classification problem with a supervised learning process.The focus of its work is on the selection of relevant features,mainly on the microblog emotion based on emotional dictionaries,user attributes and other shallow features.The deeper characteristics have not been fully explored,such as the credibility of microblog publishing sources,the emotional tendency of microblog reviews,and the structural characteristics of microblog dissemination,and thus cannot achieve a good result.Based on this,this paper analyzes the whole process from the generation of the spread of microblog rumors to the receiver of the information,and proposes a rumors recognition strategy based on the deep learning LSTM model,which can be used to identify rumors on specific topics of microblogs.The specific work is as follows:Firstly,this paper proposed a method of defining the level of credibility of microblog sources.Due to the lack of relevant and effective methods for identifying and tracing the sources of rumors,this article started with the source of Weibo rumors and built a method for defining the credibility of Weibo publishing sources according to the Weibo users' characteristics and giving the weights of the credibility to different features.Then we collect the rumor information of the specific topic from the fake information published by the Weibo community and collect its publisher information and comments and forwarding information as the rumor set,at the same time,collect the relevant information of the normal microblog as a non-rumor set,which together constitute a sample set for this experiment.Secondly,for the deep comments of Weibo's related comments and dissemination,this article uses a sentiment dictionary-based approach to obtain the emotional characteristics of the review.It constructs a tree-like structure to simulate the propagation structure of Weibo,and then uses Support Vectors Machine based on Gaussian kernel functions to train the simulation data,so as to get different propagation characteristics of rumors and non-rumors microblog,and then add the above features to the rumors recognition model to improve the accuracy of rumors recognition.Finally,because a large number of microblog rumors have obvious emotional tendencies,this paper uses the LSTM model to conduct sentiment analysis on microblog texts,and builds a Weibo rumors recognition model based on LSTM sentiment analysis.With comparing the conflicts and differences in the emotional tendency of the microblog corpora derived from the high and low credibility sources and adding the microblog comment and propagation features at the same time to identify the rumors.Experiments shown that the above methods had a better recognition effect on Weibo rumors.