Research And Implementation Of Rumor Detection Algorithm On Microblogging Platform Based On Stance Mining

Social media starts to occupy people's spare time nowadays with the advent of Web2.0 era.Microblogging platform,one of the most popular social media,has become an important information source of the society,on which millions of users produce and pick up information every day.Due to the enormous users and the fast information dissemination,rumors can spread easily through Microblogging platform and do harm to the public security.Therefore,it is necessary to find an automatic rumor detection method.Several rumor detection methods have been proposed based on machine learning.Most of the existing methods adopt feature set or deep learning to cope with this problem.However,the methods based on feature set need daunting manual effort and cannot extract high-dimensional features effectively.Although the methods based on deep learning require less manual effort,they are not accurate enough since they do not make good use of the rumor indication provided by stance information.Stance information is an important feature for rumor detection because users exposed to rumors tend to express more querying and denying stances.To make full use of the user's stance information,in this study,rumor detection models based on stance mining are proposed.First,we verifies the effectiveness of stance information based on statistics and experiment.The verification through the statistical analysis of the real data concludes that the distribution of the stance information in the rumor and non-rumor is different.Therefore,stance information could be an indication for rumor detection.We also propose a feature set including the stance features to detect rumor.Compared with the feature set without stance features,the feature set containing stance features receives better performance,which confirms the validation of stance information.After the verification,we propose a stance mining model based on CNN-GRU structure to extract the grammatical stance features and event-related stance features from the microblogging contents.Then,a microblogging rumor detection model based on the stance mining assisting task is proposed.Finally,in order to make full use of the high-value stance information,we proposes a rumor detection model with a hybrid attention mechanism to identify the high value stance information and improve the rumor detection model performance.All of these models show better performance compared with the existing models.