| 摘要: |
| 随着社交媒体平台的快速发展，虚假信息的传播已成为威胁社会稳定的重要因素。针对现有检测模型难以有效处理动态传播过程的问题，提出了一种融合异质子图序列的图卷积神经网络框架。该模型框架包含3个关键阶段：首先构建多模态推文簇数据集，采用自然语言处理技术进行语义特征提取，并构建传播路径图模型；其次构建用户-推文-单词异质图与动态传播网络，设计基于随机游走的子图抽样策略生成多尺度子图序列；最终将子图序列输入改进的图卷积神经网络进行端到端训练。实验结果表明，该模型有效解决了传播规模差异带来的特征稀疏问题，在公开数据集上的检测准确率达到92.43%，较现有假新闻图谱框架（Fake News Graph Framework，FANG）提升13.27%，召回率和F1值分别提升14.81%和15.23%。 |
| 关键词: 社交网络 假消息检测 异质图序列 图卷积神经网络（GCN） |
| DOI：10.20079/j.issn.1001-893x.250331001 |
|
| 基金项目:国家自然科学基金青年基金项目（62101095） |
|
| Fake News Detection in Social Networks via Fused Heterogeneous Subgraph Sequences |
| JIA Ying,HUANG Zhicheng,LI Jiabin,YANG Junjie,JIAO Kaiyu,HU Hangyu |
| (1.Southwest China Institute of Electronic Technology，Chengdu 610036，China; 2.School of Information and Communication Engineering，University of Electronic Science and Technology of China，Chengdu 611731，China) |
| Abstract: |
| With the rapid development of social media，fake news in social networks has increased massively，which has a bad impact on society，so fake news detection is one of the key problems that need to be solved in social networks.In order to solve the above problem，a graph convolutional neural network model based on heterogeneous graph subgraph sequences is proposed.The model has three stages.The first stage is to analyze the text attributes and propagation paths of tweets on the collected tweet cluster dataset，and use natural language processing，propagation graph construction and other means to process tweets.In the second stage，a user-tweet-word heterogeneous map and a tweet propagation network are established，and the above graphs are sampled based on random walks，and subgraph sequences are generated to represent the original graphs.Finally，the subgraph sequence is used to train the graph convolutional neural network.Experimental results demonstrate that the proposed model effectively addresses the feature sparsity caused by varying propagation scales.On public datasets，it achieves a detection accuracy of 92.43%，outperforming the Fake News Graph Framework(FANG) framework by 13.27%.The recall and F1-score are improved by 14.81% and 15.23%，respectively. |
| Key words: social network fake news detection heterogeneous subgraph sequences graph convolutional network（GCN） |
《电讯技术》编辑部 