Research on Rumor Detection Based on RoBERTa-BiGRU Model

Abstract

Rumors are false information in social media, which can have a negative impact on society. In recent years, the strong arrival of pre-training models has brought tremendous impetus to the development of natural language processing. However, the current rumor detection of deep learning does not integrate well with the pre-training model, and the model requires a lot of features. So this paper proposes a rumor detection algorithm based on the RoBERTa pre-training model and BiGRU fusion bidirectional gated recurrent unit for the Chinese data set. The method consists of two parts: (1) a highly robust transfer learning pre-training model RoBERTa, which learns contextual features of the text and vectorizes the text; (2) the BiGRU bidirectional loop gate control unit receives upstream tasks and propagates it the low-dimensional vector of the fusion feature, and output to the Soft-max two classification layer, output the prediction result. Finally, this article conducts simulaion experiments on the public data set CED_Data to detect rumors on the network text. Experimental results show that compared with existing algorithms, the method proposed in this paper can improve the accuracy of rumor detection on this data set.