耿唯佳, 宋玉蓉, 周伟伟. 融合TextCNN与TextRNN模型的谣言识别方法[J]. 微电子学与计算机, 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 引用本文: 耿唯佳, 宋玉蓉, 周伟伟. 融合TextCNN与TextRNN模型的谣言识别方法[J]. 微电子学与计算机, 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 GENG Weijia, SONG Yurong, ZHOU Weiwei. Rumor recognition method combining TextCNN and TextRNN[J]. Microelectronics & Computer, 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 Citation: GENG Weijia, SONG Yurong, ZHOU Weiwei. Rumor recognition method combining TextCNN and TextRNN[J]. Microelectronics & Computer , 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 耿唯佳, 宋玉蓉, 周伟伟. 融合TextCNN与TextRNN模型的谣言识别方法[J]. 微电子学与计算机, 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 引用本文: 耿唯佳, 宋玉蓉, 周伟伟. 融合TextCNN与TextRNN模型的谣言识别方法[J]. 微电子学与计算机, 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 GENG Weijia, SONG Yurong, ZHOU Weiwei. Rumor recognition method combining TextCNN and TextRNN[J]. Microelectronics & Computer, 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672 Citation: GENG Weijia, SONG Yurong, ZHOU Weiwei. Rumor recognition method combining TextCNN and TextRNN[J]. Microelectronics & Computer , 2022, 39(1): 31-38. doi: 10.19304/J.ISSN1000-7180.2021.0672

传统的谣言识别方法耗费人力物力并且准确率较低。为了有效识别社交网络中的谣言，提出一种基于融合模型的谣言识别方法.该方法首先通过BERT预训练模型构建文本句向量；其次构建TextCNN模型挖掘文本的语义特征，构建TextRNN模型用于挖掘文本的时序特征；最后，对两种模型进行加权融合，实现对谣言的识别.此外，还对原始主流模型进行了改进，一是借鉴Inception模型的思想来增加TextCNN模型的深度，二是将注意力机制注入TextRNN模型中，增加其可解释性和泛化能力.实验结果表明，相较于当前主流的谣言识别方法，该方法准确率可达到97.12%并且F1值可达到97.14%.

社交网络 /  谣言识别 /  BERT模型 /  文本卷积神经网络 /  文本循环神经网络 Abstract:

The traditional rumor recognition method consumes manpower and material resources and has a low accuracy rate. In order to effectively identify rumors in social networks, a rumor recognition method based on a fusion model is proposed.Firstly, the text sentence vector is constructed through the BERT pre-training model. Secondly, the TextCNN model is constructed to mine the semantic features of the text, and the TextRNN model is constructed to mine the temporal features of the text. Finally, the two models are weighted and fused to realize the identification of rumors. In addition, the original mainstream model has been improved, One is to learn from the idea of the Inception model to increase the depth of the TextCNN model, and the other is to inject the attention mechanism into the TextRNN model to increase its interpretability and generalization capabilities. The experimental results show that compared with the current mainstream rumor recognition method, the accuracy of this method can reach 97.12% and the F1 value can reach 97.14%.

Key words: social network /  rumor detection /  BERT model /  text convolutional neural network /  text recurrent neural network  高艺璇. 突发公共卫生事件中网络谣言传播探析——以新冠肺炎疫情相关谣言为例[J]. 新闻研究导刊, 2020, 11(7): 10. DOI: 10.3969/j.issn.1674-888 3.2020.07.006.

GAO Y X. Analysis on the spread of online rumors in public health emergencies: taking rumors related to the new crown pneumonia epidemic as an example[J]. News Research Guide, 2020, 11(7): 10. DOI: 10.3969/j.issn.1674-8883.2020.07.006 . ZHOU K M, SHU C, LI B Y, et al. Early rumour detection[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Minneapolis, Minnesota: Association for Computational Linguistics, 2019: 1614-1623. DOI: 10.18653/v1/N19-1163 . DEVLIN J, CHANG M W, LEE K, et al. BERT: pre-training of deep bidirectional transformers for language understanding[C]//Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Minneapolis, Minnesota: Association for Computational Linguistics, 2019: 4171-4186. DOI: 10.18653/v1/N19-1423 . PETERS M E, NEUMANN M, IYYER M, et al. Deep contextualized word representations[C]//Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. New Orleans, Louisiana: Association for Computational Linguistics, 2018: 2227-2237. DOI: 10.18653/v1/N18-1202 .