| 注册
首页|期刊导航|情报杂志|融合响应特征差异的多模态AI虚假信息检测

融合响应特征差异的多模态AI虚假信息检测

张新生 林承宇 马玉龙 王润周

情报杂志2025,Vol.44Issue(7):122-134,13.
情报杂志2025,Vol.44Issue(7):122-134,13.DOI:10.3969/j.issn.1002-1965.2025.07.016

融合响应特征差异的多模态AI虚假信息检测

Multimodal AI Disinformation Detection Based on Differences in Response Features

张新生 1林承宇 1马玉龙 1王润周1

作者信息

  • 1. 西安建筑科技大学管理学院 西安 710055
  • 折叠

摘要

Abstract

[Research purpose]Criminals use AI to generate false graphical information,disrupt public perception,and threaten social stability.The existing detection methods are lack generality in the face of information generated by different AI models,and it is difficult to accurately identify false content.In order to solve this challenge,this study explores an effective detection mechanism based on the differences in response features of generative AI to improve the accuracy and robustness of disinformation detection.[Research method]Through the comparative analysis of the text features of GPT,PaLM2,Llama2,Mixtral and RWKV,the multi-dimensional difference features such as lexical parts of speech,emotional tendencies,grammatical structure,and confusion were extracted,and the BVT-CNN multimodal disinformation detection model fused with BERT and Vision-Transformer was introduced.The ablation comparison method was used to evaluate the influence of differential features on the detection performance.[Research result/conclusion]The results show that the wF1 index increased by 4.94%after the fusion of differential features,which significantly enhances the detection ability of hybrid generative AI information.The research results not only optimize the AI disinformation detection strategy,but also summarize and analyze the differences in response features of various generative models.

关键词

生成式AI/虚假信息检测/响应特征/普适性检测/差异特征/多模态

Key words

generative AI/disinformation detection/response features/universal testing/difference features/multimodal

分类

社会科学

引用本文复制引用

张新生,林承宇,马玉龙,王润周..融合响应特征差异的多模态AI虚假信息检测[J].情报杂志,2025,44(7):122-134,13.

基金项目

教育部人文社会科学规划基金项目"泛在信息社会下AI生成式虚假信息风险感知及治理路径研究"(编号:24YJA630129) (编号:24YJA630129)

陕西省社会科学基金年度项目"AIGC时代下生成式虚假信息风险感知及治理路径研究"(编号:2024R083) (编号:2024R083)

陕西省自然科学基础研究计划项目"AIGC背景下虚假信息演化、识别及治理研究"(编号:2025JC-YBMS-1100)研究成果. (编号:2025JC-YBMS-1100)

情报杂志

OA北大核心

1002-1965

访问量0
|
下载量0
段落导航相关论文