The Research And Implementation Of Rumor Detection Method On Social Media

In the era of mobile internet,social media is integrated into people’s life,and has become an important platform for information obtain and change.However,while social media brings convenience to people,it also facilitates the generation and dissemination of rumors.Recently,the rumors on social media have misled public perception,which caused adverse impact on the individual life and social stability.Faced with the huge amount of information on social media,it is not realistic to identify rumors manually.Therefore,how to detect social media rumors effectively with automated technologies and models is crucial to the ecological governance of network.Traditional deep models can automatically extract features of the rumor,but cannot effectively mine propagation relationships of the rumor.Although the existing methods of rumor detection using graph neural networks can solve this problem,these methods focus on modeling the final structure of information propagation,while ignoring the dynamic evolution process of the propagation structure over time.Currently,the success of deep learningbased rumor detection methods relies on largescale annotated data,but it is difficult to obtain sufficient annotated data about the new events.In responses to these shortcomings,the paper aims to improve the existing methods to improve the performance of rumor detection on social media.The main research achievements of the paper are as follows:Aiming at the problems existing in the current rumor detection methods which use graph neural networks to mine propagation relationships,this paper proposes a model which integrates temporal dynamic structure and emotional information for rumor detection on social media.(1)We construct a dynamic propagation graph based on the post time of rumor tweets and their responsive tweets,and model the information propagation structure in different time stages using modified graph convolutional network;(2)We use attention mechanismbased gated recurrent units to extract temporal dynamic structure representation.Beside we extract sentiment representation using pretrained language model and selfattention mechanism,and concatenate the above representations for detecting rumor on social media?(3)We conduct experimental verification on Sina Weibo rumor detection dataset.The results demonstrate the superiority and interpretability of the proposed model in this paper.Aiming at the problems existing in detecting the rumor on newly emerged events with the acquisition of annotation data,this paper proposes a fewshot metalearning method for rumor detection based on event semantic similarity.(1)We first split the instances based on events,and then use pretrained language model and gated recurrent units to extract content representations of original tweets.Beside we manually extract user features and extract potential interaction representations among users using graph convolutional networks;(2)We design a fewshot metalearning method for rumor detection based on event semantic similarity to help the model obtain better initial knowledge from the rumor detection task of historical events,which can benefit the detection of rumor on newly emerged events.(3)We conduct comprehensive experiments on two social media rumor detection datasets.The results validate that the accuracy of fewshot rumor detection can be improved through utilizing the similarity between events.