Rumor Detection Algorithm And Application Based On Deep Learning

In recent years,the Internet has developed rapidly,and social networks have become more and more important in people's lives,such as Weibo,Twitter,QQ,We Chat,Facebook and so on.Social networks can share and disseminate real-time news,and the way of information transmission has changed dramatically.However,the information transmitted by users in social networks may not be true.In social networks,rumors spread faster and more widely,so the harm caused is more huge.In view of the fact that online rumors have seriously affected people's lives and social stability,all sectors of government and social networking platforms are paying more and more attention to how to identify rumors in a timely,rapid and accurate manner through relevant technical means.At present,the mainstream rumor identification method of social network platform is manual identification,which consumes manpower and material resources and has low efficiency.Therefore,many researchers have proposed various rumor recognition methods based on machine learning model or deep learning model.These methods have good effects,but they also have some shortcomings.The research purpose of this topic is to take Sina Weibo as the representative,to mine all kinds of features from the communication information of Weibo social network,and to improve and integrate several mainstream deep learning models to realize the recognition of rumors.The main research content and achievements of this paper are as follows::(1)The Bert model,which has a good effect at present,is used as the natural language pre-training model to transform the text of micro-blog into vector representation.Then the improved text convolutional neural network I-CNN model and the improved text cyclic neural network I-RNN model are constructed.At the same time,we have done many experiments on the input vector dimension of i-CNN model,which shows that it has some advantages in early rumor recognition.(2)Because the I-CNN model can extract semantic information of text and capture contextual relevance information,the I-RNN model has great advantages for feature mining of sequences.Therefore,this paper uses a feasible fusion model technique to combine the two improved models in(1).At the same time,some traditional features,such as user characteristics,time characteristics and propagation characteristics,were extracted manually from the data set and combined with the deep learning model to improve the prediction and generalization ability of the model.Finally,the results are compared with those of several papers using the same dataset.Experiments show that this model can accurately identify rumors,and is superior to other models in accuracy and F1 value.(3)Using the model constructed in the method(2),develop a rumor detection system based on Weibo OAuth.First obtain account information and Weibo comment content from the official Weibo interface,then call the trained model to calculate the probability that Weibo is a rumor,and finally pass all the information to the user.