基于BERT-RCNN模型的微博谣言早期检测研究

Early Detection of Micro Blog Rumors Based on BERT-RCNN Model

[目的/意义]为了解决传统谣言检测算法在实际应用中存在滞后性的问题,尝试在不使用评论和转发数据的基础上实现微博谣言早期检测。[方法/过程]针对传统谣言检测模型需要大量特征以及难以实现及时检测的问题,使用BERT模型对微博原文进行向量表示,然后将获取的语义特征输入到RCNN模型中进行谣言检测。[结果/结论]基于微博谣言数据集进行对比实验,实验结果显示基于BERT-RCNN的微博谣言早期检测模型的准确率为95.16%,F1指标为95.14%;该模型与其他主流的谣言检测模型相比,能在较短时间内完成对谣言的检测,证明了方法的有效性。[局限]文章提出的检测模型针对微博文本进行检测,BERT模型需要依赖谷歌发布的预训练模型。

[Purpose/significance]In order to solve the problem of lag in the actual application of traditional rumor detection algorithms,an attempt was made to achieve early detection of microblog rumors without using comments and forwarding data.[Method/process]In view of the problems that traditional rumor detection models require a large number of features and are difficult to achieve timely detection,the BERT model is used to vectorize the original text of Weibo,and then the acquired semantic features are input into the RCNN model for rumor detection.[Result/conclusion]A comparative experiment was carried out on the Weibo rumors dataset.The experimental results proved that the accuracy of the early detection model of Weibo rumors based on BERT-RCNN was 95.16%,and the F1 index was 95.14%.[Limitations]The detection model proposed in this article is for microblog text detection,and the BERT model needs to rely on the pre-training model released by Google.