情报杂志2025，Vol.44Issue(11)：188-197,135,11.DOI:10.3969/j.issn.1002-1965.2025.11.023
基于扩散模型数据增强的突发公共卫生事件谣言检测研究
Rumor Detection of Public Health Emergencies Using Diffusion Model-Based Data Augmentation
摘要
Abstract
[Research purpose]To address the issues of data scarcity and imbalanced distribution in the task of rumor detection during pub-lic health emergencies,this paper introduces a diffusion model to balance the diversity and semantic consistency of generated samples in da-ta augmentation methods.[Research method]A diffusion model-based data augmentation approach for rumor detection in public health emergencies is proposed,comprising three core components:a sample generation conditional mask module,a diffusion model data aug-mentation module,and a rumor detection module.The sample generation conditional mask module generates semantically guided rumor mask texts by combining dynamic noise scheduling and label-aware embedding.The diffusion model data augmentation module produces pseudo-samples with both distribution rationality and semantic consistency through forward perturbation and reverse restoration.The rumor detection module achieves efficient utilization of augmented data by integrating a similarity selection weighting mechanism and a contrastive learning-driven noise adversarial training strategy.[Research result/conclusion]Experimental results demonstrate that our method a-chieves superior accuracy and F1-scores across Chinese,English,Arabic,and German datasets compared to various augmentation-based and non-augmented baselines,with performance gains ranging from 0.2％to 2.6％,surpassing the current state-of-the-art baseline models.The findings demonstrate that integrating diffusion model-based data augmentation with contrastive learning advantages can signif-icantly enhance rumor detection performance in public health emergencies.关键词
突发公共卫生事件/网络谣言/舆情监测/谣言检测/扩散模型/数据增强Key words
public health emergencies/online rumors/public opinion monitoring/rumor detection/diffusion models/data augmentation分类
计算机与自动化引用本文复制引用
王连喜,廖信峰,陈宣齐,陈卓玮,容梓莹..基于扩散模型数据增强的突发公共卫生事件谣言检测研究[J].情报杂志,2025,44(11):188-197,135,11.基金项目
国家社会科学基金项目"突发公共卫生事件舆情中网民负面情绪检测及引导研究"(编号:22BTQ045)研究成果. （编号:22BTQ045）
