Research On Rumor Detection Technology In Social Network Based On Deep Learning

With the rapid development of Internet technology,social networking platforms such as Weibo and Twitter have become the main platforms for netizens to express their opinions and obtain information.Due to the fast speed and low cost of information dissemination in social networks,the identity of netizens has the characteristics of virtualization and concealment,which provides convenient conditions for the breeding and dissemination of online rumors.Rumors on the Internet are especially harmful to the security and stability of the country and society.Using artificial methods to identify network rumors has problems such as high cost and low efficiency.With the rapid development of artificial intelligence technology represented by deep learning,the field of intelligent rumors detection in social networks has begun to become a research hotspot.Starting from the above background,this thesis studies the intelligent rumor detection technology of social networks based on deep learning,mainly from two aspects:rumor veracity determining and early rumor detection.Because the original tweets and reply messages in Twitter have a large difference in the expression characteristics of the content,and the content of both also has the problem of data redundancy,which affects the accuracy of the veracity of the rumor.In response to the above problems,this thesis proposes a dual-engine rumor veracity determining model based on deep learning to realize the differential processing of the feature extraction of original tweets and reply information;in addition,this thesis proposes a DSA(Double Self-Attention)mechanism,from two dimensions: sentences and words,to simultaneously eliminate data redundancy.The experimental results had shown that the rumor veracity determining model proposed in this thesis has a higher accuracy rate.Reinforcement learning technology is one of the effective ways to solve the problem of early detection of online rumors.However,in the background of early rumors detection,current reinforcement learning models have problems such as ignoring the potential meaning of state sequences and imperfection in reward functions.In addition,current rumor detection model for early detection also has the problem of lacking in processing information.In response to the above problems,this thesis proposes an early rumor detection model based on DRQN(Deep Recurrent Q-Learning Network),which uses LSTM(Long Short-Term Memory)to learn state sequence features,and optimizes the reward function from the aspects of the timeliness and accuracy of rumor detection.Also,this thesis uses the DSA-based dual-engine model to improve the current rumor detection model.Experiments had shown that the early rumor detection model proposed in this thesis has a great improvement in the timeliness and detection rate of rumor detection.