Research On Microblog-oriented Rumor Detection And Analysis

Sina micro-blog as a typical representative of Web2.0 applications,provides a platform of information dissemination and exchange for micro-blog users,with the information content of openness,brevity and information exchange of convenience.However,due to the lack of effective supervision and verification of contents,plenty of rumors exist in micro-blog and widely spread,give rise to damaging the interests of individuals,causing social panic.Therefore,research on microblog-oriented rumor detection and analysis contributes to realize the automation of detecting rumors,early detection of rumors and prevent its spread,provides assistance for the departments to control and govern rumors,and has an important realistic significance to build a healthy network environment.The thesis makes research on microblog-oriented rumor detection and analysis.The main contributions are listed as follows:(1)The existing methods for rumor detection only select shallow content features and information propagation features,and does not dig out deep micro-blog content features and propagation features.In addition,the existing studies use a single classifier to detect micro-blog rumors without combining multiple classifiers to ensemble classifier.To solve this problem,a method based on deep features and ensemble classifier is proposed.Firstly,features are extracted from the sentiment orientation,propagation and user's historical information of micro-blog.Then,the classifier is trained by using these deep features.Finally,micro-blog rumors are detected based on the ensemble classifier.Experimental results on micro-blog corpus show that the method can effectively improve the classification performance.(2)Due to text-picture unmatched rumors,a method base on entity linking is proposed.Firstly,text-picture match degree feature is computing by entity linking.Then,in order to verify the features extracted from the sentiment orientation,propagation and user's historical information of micro-blog are valid to detect text-picture unmatched rumors.The classifier is retrained by using text-picture match degree feature and these four classifications.Finally,micro-blog rumors are detected based on the classifier.Experimental results on micro-blog corpus show that the proposed text-picture match degree feature can effectively improve the precision ratio of rumor detection.(3)The system of micro-blog rumors detection and analysis is developed to show the micro-blog user's personal information,micro-blog content,content of micro-blog additional picture and micro-blog propagation process.Micro-blog credibility is judged by the analysis of various aspects of the micro-blog features.