基于潜在话题的微博谣言在线检测

Online Detection of Weibo Rumors Based on Potential Topics

微博谣言检测对净化社交网络环境与维护社会和谐发展有重要意义。主流的社会网络谣言自动检测方法甄别网络中每一条信息的真实性,导致现有方法存在检测准确率低,检测试时延大的问题。针对以上问题,提出一种基于潜在话题的微博谣言检测,该模型借鉴热力学中的热量传播模型提出一种改进的基于热量模型的微博潜在话题检测方法,该模型首先通过改进的热量模型提取出潜在话题的微博,解决在现有的微博谣言检测工作中,无法在海量的微博中进行谣言检测而面临的谣言检测的冷启动问题。然后使用LDA主题模型提取出具有相同主题特征的微博归为一类进行检测,从而解决对单个微博检测而忽略谣言之间的联系。最后将微博输入到Transformer模型中,对微博的内容进行深层语义挖掘与分析,从而识别谣言。实验结果表明,所提出的方法在谣言早期检测方面较本文设置的表现最好基线方法提前1.5小时,在检测精准度上也有较好的性能。

The detection of rumors on Weibo is of great significance for purifying the social network environment and maintaining the harmonious development of society.The mainstream automatic detection method of social network rumors discriminates the authenticity of each piece of information in the network,resulting in the problems of low detection accuracy and large test delay in the existing methods.Regarding the issue above.For the above problems.Proposes a microblog rumors detection based on latent topics.The model draws on the heat propagation model in thermodynamics to propose an improved microblog latent topic detection method based on thermal models.The model is first extracted by the improved thermal model The potential topic of microblogging solves the problem of cold start of rumors detection in the existing microblogging rumors detection work,which cannot be detected in massive microblogs.Then use the LDA topic model to extract the microblogs with the same topic features into one class for detection,thus solving the detection of a single microblog and ignoring the connection between rumors.Finally,the microblog is input into the Transformer model,and the content of the microblog is deeply semantically mined and analyzed to identify rumors.Experimental results show that the method proposed in this paper is 1.5 hours ahead of the best performing baseline method set in this paper in the early detection of rumors,and it also has better performance in detection accuracy.