Research And Implementation Of Early Rumor Detection Based On Hierarchical Attention Network

In recent years,with the rise of social media,the problems caused by rumors have become more serious than ever.Because rumors involve public topics such as economics,health,and politics,their authenticity and correctness cannot be verified on time or they can never be verified.These rumors carry unconfirmed or even false information,which may cause public panic,resulting in serious economic losses and adverse effects on society.Faced with the huge amount of information on social media,the time and effort required by professionals to identify rumors are huge,and there are problems such as delayed recognition and incomplete coverage.Therefore,it is of great significance to research and design an automatic identification method of rumors that has both accuracy,coverage,and timeliness.It can promptly remove and delete rumors before they form an effective scale,reduce subsequent adverse effects,and prevent rumors before they occur.Based on this purpose,this research uses WeChat rumor datasets we collect,combined with hierarchical attention networks,deep reinforcement learning,and generative adversarial methods to design and implement hybrid rumor detection algorithms and early rumor detection algorithms.The main work of this article is as follows:(1)Collect and construct a WeChat rumor data set,and preprocess and analyze it.Specifically,first,we crawled WeChat public account articles through web crawlers to obtain a large amount of text and social context information of rumor events and real events.Secondly,for the subsequent analysis and modeling tasks,three pre-processing operations are performed:cleaning,Chinese word segmentation,and stop words removal.Then,we analyze the differences in distribution between rumors and non-rumors from the perspective of timeliness and text content and discuss the reasons for presenting these differences.(2)A hybrid rumor detection method based on a hierarchical attention network is designed.We propose an HHAN(Hybrid Hierarchical Attention Network)model,which combines the hierarchical semantic information and temporal information of events on social media;also,it extracts time-based and content-based statistical features and integrates them into the model.Through many experiments on the WeChat rumor data set,the effectiveness of our proposed HHAN model on rumor detection tasks is verified.(3)An early rumor detection method combining hierarchical attention network and reinforcement learning technology is designed.We propose a HAN-ERD(Hierarchical Attention Network-based Early Rumor Detection)model is proposed.This model uses hierarchical attention networks to model hierarchical semantic information and time-series information to ensure good detection results;on the other hand,it uses deep reinforcement learning technology to implement the checkpoint module so that it can learn the number of posts required to trigger the detection module and achieve the effect of early detection.Furthermore,using the generative adversarial learning technology and combining the first two modules,the HAN-ERD model can not only maintain good detection performance but also ensure the timeliness of detection.Finally,we conducted experiments on the WeChat rumor dataset to verify the effectiveness of the HAN-ERD method on early rumor detection tasks.