Research On Scientific Rumor Detection And Rumor Tracking On Online Social Networks Based On Deep Learning

With the development of Internet infrastructure and mobile Internet technology,online social media is in the ascendant.Massive information on social media contains many unconfirmed or confirmed rumors that are not known to all users.These rumors may cause improper attention or mislead people’s lives.This thesis divides rumors on online social networks into two categories: scientific rumors and social rumors.Scientific rumor refers to the rumor that relevant professional knowledge is needed in clarification.Otherwise,rumor is attributed to social rumor.A complete rumor confrontation system includes four stages: rumor detection,rumor tracking,stance classification,and rumor veracity.In view of the problems such as the use of technical means for scientific rumor detection,the lack of scientific rumor data limits the ability of the detection model and rumor tracking,this thesis conducts in-depth exploration.The main work of this thesis is as follows :1)This thesis proposes the relative influence index of scientific rumors to quantify the influence of scientific rumors on online social networks,proving the necessity of using technical means to detect scientific rumors.Quantifying the influence of scientific rumors and exploring the use of technical means to detect scientific rumors is still a blank field.2)This thesis constructs a data set for scientific rumor detection,including three types of data: scientific rumor,social rumor,and non-rumor text.This is the first dataset for scientific rumor detection to the best of the authors’ knowledge.3)A model based on long short-term memory(LSTM)is proposed for scientific rumor detection.The work of this thesis can provide a benchmark for the detection of scientific rumors on online social networks,and the model proposed in this thesis shows excellent competitiveness compared with multiple baseline models.4)In view of the lack of scientific rumors data,this thesis proposes to detect scientific rumors from the perspective of generating scientific rumors.Therefore,this thesis presents a scientific rumor generation model based on the Transformer and LSTM models and uses the generated data to improve the ability of the detection model.The model proposed in this thesis can sample and generate semantically coherent scientific rumors in the vocabulary space,and the generated data has a significant boost for all detection models.Furthermore,our model escapes the limitation of edit-based models that specify the number of edit times.5)This thesis proposes the Sim CLRT model for solving the rumor tracking task based on contrastive learning and the BERT model.Sim CLRT contains three variants--Sim CLRT-CNN,Sim CLRT-RNN,Sim CLRT-Linear.Sim CLRT can track events with more tweets and devote enough attention to the events related to fewer tweets.Sim CLRT achieves state-of-the-art results on the two general rumor tracking datasets used in this thesis.