Research And Implementation Of Rumor Detection Based On Comment Structure

The rapid development of social media has changed the way people communicate with each other in daily life,but it has also led to the creation of rumors.Rumors spread quickly and widely,and the proliferation of rumors can seriously pollute the social network ecology and affect users’ access to high-quality information.Therefore,rumor detection has become an important research task in academia and industry.The research focus on rumor detection mainly includes two aspects: timeliness(early detection)and accuracy(precise detection).However,existing rumor detection methods face challenges in both aspects: first,although the existing early rumor detection methods can extract the structural features of rumor propagation,they do not take into account the characteristic of dynamic propagation,so it is difficult to identify rumors in a timely manner.Second,it is necessary to further excavate its details for rumor.However,the rumor detection methods based on graph neural networks cannot effectively mine the deep-level properties of rumors and it is difficult to extract the dynamic interactive features of comment structure evolution over time,which limits the improvement of accuracy.To address the above problems,this thesis takes into account the timeliness and accuracy of rumor detection.By modeling the structural relationship between comments in the process of rumor propagation,this thesis deeply investigates new rumor detection methods with early detection capability and accurate detection capability,respectively.First,this thesis proposes Early Rumor Detection Based On Stance Feature and Reinforcement Learning(ERD-SFRL)to detect rumors in a timely manner in the process of rumor propagation.Then,this thesis proposes Rumor Detection Based On Dynamic Graph Attention Capsule Network(DYN-GACN)to improve the accuracy of rumor detection by mining the deep-level properties of rumors and the dynamic interaction characteristics in the dynamic evolution of comment structure over time.Finally,based on the above two algorithms,this thesis designs and implements a proto-system for social media rumor detection,and conducts detailed functional tests.The main work of this thesis is as follows:(1)In order to solve the problem that existing rumor detection methods do not sufficiently take into account the characteristic of dynamic propagation,thus making it difficult to detect rumors in a timely manner,this thesis proposes Early Rumor Detection Based On Stance Feature and Reinforcement Learning(ERD-SFRL).Specifically,ERD-SFRL consists of three components: a stance-aware module,which acts as a model to extract stance features from comments,a rumor classification module which is used for rumor identification by dividing the rumor propagation process into a series of sub-comment structures based on graph structures and a reinforcement decision module is applied to select the appropriate action for early detection.Moreover,both the rumor classification module and the reinforcement decision module incorporate stance features to enhance the feature representation.(2)Aiming at mining the deep-level properties of rumors and the dynamic interaction features of comment structure evolution over time that improves the accuracy of rumor detection,this thesis proposes Rumor Detection Based On Dynamic Graph Attention Capsule Network(DYNGACN).Specifically,DYN-GACN consists of two components: a module(DYN)that can divide the comments accumulated in the process of rumor propagation in chronological order to form multiple static sub-comment structures and a graph attention network module(GACN),incorporating the source post features,can encode static structures into substructure classification capsules for mining the deeper properties of rumor.Moreover,a classification capsule attention mechanism is designed to focus on the important information of each substructure classification capsule,so as to capture the dynamic interaction features during the dynamic evolution of the rumor comment structure over time.(3)Based on the above two algorithms,this thesis designs and implements a proto-system for social media rumor detection.The whole system consists of four main functions,including a data collection module,a rumor classification module,a historical review of rumor posts,and a visual analysis,and uses the Uniform Content Label(UCL)to unify rumor-related data.In the rumor classification module,ERD-SFRL and DYN-GCAN are used to provide users with timely and accurate detection effects,thus improving users’ ability to identify rumors.Finally,the usability and robustness of the proto-system are verified by functional testing of the system.