基于混合神经网络的突发公共卫生事件微博谣言识别研究  

Research on Weibo Rumor Identification in Public Health Emergencies Based on Hybrid Neural Network

作　　者：冯兰萍[1,3] 董陈超 徐绪堪 Feng Lanping;Dong Chenchao;Xu Xukan(Business School,Hohai University,Changzhou 213022;School of Economics and Management,Southeast University,Nanjing 211189;Changzhou Key Laboratory of Industrial Big Data Mining and Knowledge Management,Changzhou 213022)

机构地区：[1]河海大学商学院,常州213022 [2]东南大学经济管理学院,南京211189 [3]常州工业大数据挖掘与知识管理重点实验室,常州213022

出　　处：《情报杂志》2022年第12期81-88,共8页Journal of Intelligence

基　　金：国家社会科学重大基金项目“总体国家安全观下重大突发事件的智能决策情报体系研究”(编号:20&ZD125)子课题“总体国家安全观下重大突发事件智能决策情报体系的基本理论与方法研究”研究成果之一。

摘　　要：[研究目的]提高突发公共卫生事件中微博谣言识别效果,分析微博特征对谣言识别的影响,为公众识别谣言、政府及相关部门干预治理提供参考。[研究方法]在用户基本特征和内容特征、传播特征的基础上引入用户历史特征,构建基于混合神经网络的多特征融合谣言识别模型,该模型采用BiLSTM+CNN抽取代表性深度语义特征,与其他特征融合后,使用DNN网络进行谣言识别。[研究结论]实验结果表明:该模型的准确率、查准率、查全率、F1值均高于94%,准确率、查全率、F1值优于其他谣言识别模型;语义特征是谣言识别的关键特征,在此基础上融合单一特征能够大幅提高谣言识别效果,提高程度由高到低,依次为统计特征、传播特征、用户基本特征、用户历史特征和情感特征。递进融合特征越多,模型平均性能、泛化能力越强,谣言识别效果越好。[Research purpose]To improve the identification effectiveness of Weibo rumor in public health emergencies,and to analyze the influence of Weibo features on the identification of rumor,so as to provide support to the public to identify rumors and the government and relevant departments to intervene in the governance of rumors.[Research method]Based on the user basic features,the content features,and the spreading features,user historical features are introduced,a multi-feature fusion rumor identification model is constructed based on hybrid neural network.This model use BiLSTM+CNN to extract the representative depth semantic features and then merge with other features,use DNN network for rumor identification.[Research conclusion]The experimental results show that the accuracy,precision,recall and F1 value of the model are all higher than 94%,the accuracy,recall and F1 value are obviously superior to other rumor identification models.The fusion of semantic features and other features can greatly improve the effectiveness of rumor identification.From high to low in the improvement degree of rumor identification,they are statistical features,spreading features,user basic features,user historical features,and emotional features in turn.The more incremental features,the better the average performance and generalization ability of the model,and the better the rumor identification.

关 键 词：公共卫生事件 微博谣言 谣言识别 多特征融合 混合神经网络 

分 类 号：TP393[自动化与计算机技术—计算机应用技术]