首页|融合用户特征的图注意力微博谣言检测模型

融合用户特征的图注意力微博谣言检测模型

扫码查看
随着网络和通信技术的发展,谣言借助微博等平台可快速扩散,形成病毒式传播,给国家安全和社会稳定造成严重的安全隐患.为了提高谣言自动检测的准确率,对基于图注意力网络的全局-局部注意力编码谣言检测模型进行了改进.首先,引入用户属性信息对微博文本内容特征和传播结构特征进行补充,得到更高阶特征;其次,改进图注意力机制以得到更健壮的节点聚合特征,为判决是否为谣言提供更准确的依据.在微博谣言数据集上的实验结果表明,相对于已有算法,该文提出的检测模型具有更高的检测准确率.
Incorporating User Features for Weibo Rumor Detection via Graph Attention Network
With the development of network and communication technology,rumors spread rapidly like virus with in Weibo and other platforms,causing serious risks to national security and social stability.In order to improve the ac-curacy of automatic rumor detection,we present a global-local attention network for rumor detection model.Firstly,user features are introduced as higher-order features in addition to the text information and propagation structure features of Weibo.Secondly,the graph attention network is improved to obtain more robust node aggregation fea-tures,which provides more accurate evidence for judging a rumor.The experimental results on the Weibo rumor dataset show that the detection model proposed in this paper has higher accuracy than the existing algorithms.

rumor detectiongraph attention networkuser featuresstructure propagation information

杨帆、李邵梅

展开 >

解放军战略支援部队信息工程大学,河南郑州 450001

谣言检测 图注意力机制 用户属性信息 传播结构信息

国家自然科学基金

61521003

2024

中文信息学报
中国中文信息学会,中国科学院软件研究所

中文信息学报

CSTPCDCHSSCD北大核心
影响因子:0.8
ISSN:1003-0077
年,卷(期):2024.38(8)