ThinkDeception: A Progressive Reinforcement Learning Framework for Interpretable Multimodal Deception Detection
作者: Jinhao Song, Shan Liang, Yiqun Yue, Zhuhuayang Zhang, Tianqi Gao
分类: cs.AI
发布日期: 2026-06-17
备注: 10pages,4figures
💡 一句话要点
提出ThinkDeception以解决多模态欺骗检测的可解释性问题
🎯 匹配领域: 支柱二:RL算法与架构 (RL & Architecture) 支柱九:具身大模型 (Embodied Foundation Models)
关键词: 多模态欺骗检测 可解释性 认知推理 多模态大型语言模型 渐进训练策略 模态不一致性 思维链数据集
📋 核心要点
- 现有的多模态欺骗检测方法主要依赖黑箱模型,缺乏可解释性,难以捕捉细微的跨模态不一致性。
- 本文提出ThinkDeception框架,将多模态大型语言模型引入欺骗检测,转变为认知推理过程,并利用逐步思维链数据集进行训练。
- 实验结果表明,ThinkDeception在检测准确性和推理质量上显著优于现有方法,建立了新的SOTA。
📝 摘要(中文)
多模态欺骗检测对于识别欺诈意图至关重要,但现有方法主要依赖于端到端的黑箱模型,缺乏可解释性,无法提供透明的推理过程,且难以捕捉欺骗行为中的细微跨模态不一致性。为了解决这些问题,本文提出了ThinkDeception,一个新颖且可解释的多模态欺骗检测框架。该框架首次将多模态大型语言模型(MLLMs)引入该领域,将欺骗检测从传统的二分类任务转变为明确的认知推理过程。通过精心注释的逐步多模态思维链(CoT)数据集,开发了基础模型ThinkDeception Base,验证了模态不一致性在解码欺骗中的关键作用。我们的核心创新在于提出了视觉-音频一致性组相对策略优化(VAC-GRPO),并配备渐进训练策略,显著提升了模型的推理质量。
🔬 方法详解
问题定义:本文旨在解决多模态欺骗检测中的可解释性不足问题。现有方法多为黑箱模型,无法提供透明的推理过程,且难以捕捉欺骗行为中的细微跨模态不一致性。
核心思路:提出ThinkDeception框架,通过引入多模态大型语言模型(MLLMs),将欺骗检测转变为认知推理过程,增强模型的可解释性和推理能力。
技术框架:整体架构包括基础模型ThinkDeception Base和VAC-GRPO模块,前者通过逐步思维链数据集进行训练,后者则采用渐进训练策略,分为四个难度层次。
关键创新:VAC-GRPO是本文的核心创新,通过动态课程调度器和多维过程感知奖励机制,显著提升模型的推理质量,与传统GRPO方法相比,提供了更为系统的训练策略。
关键设计:模型设计中采用了分层训练数据、反思学习范式,以及针对模态不一致性的奖励机制,确保模型在不同难度层次上逐步提升推理能力。
🖼️ 关键图片
📊 实验亮点
在实验中,ThinkDeception在多个主流基准上建立了新的SOTA,检测准确性显著提高,推理质量也得到了显著改善。具体而言,模型在准确性上提升了XX%,在推理质量上超越了现有方法,展示了其在多模态欺骗检测中的优越性。
🎯 应用场景
该研究具有广泛的应用潜力,尤其在金融欺诈检测、网络安全、社交媒体监控等领域。通过提供可解释的欺骗检测机制,能够帮助相关行业更有效地识别和应对欺诈行为,提升决策的透明度和准确性。未来,该框架还可能扩展到其他需要多模态分析的领域,如心理学研究和人机交互。
📄 摘要(原文)
Multimodal deception detection is critical for identifying fraudulent intentions, yet existing approaches predominantly rely on end to end black--box paradigms. These methods suffer from a severe lack of interpretability failing to provide transparent reasoning trajectories and struggling to explicitly capture the subtle, cross modal inconsistencies inherent in deceptive behaviors. To transcend these limitations, we propose ThinkDeception, a novel and interpretable multimodal deception detection framework. As a pioneering effort, it introduces Multimodal Large Language Models (MLLMs) into this domain, transforming deception detection from a traditional binary classification task into an explicit cognitive reasoning process. Facilitated by the first meticulously annotated step--by--step multimodal Chain of Thought (CoT) dataset, we develop a foundational model, ThinkDeception Base, empirically validating the critical role of modal inconsistency in decoding deception. Building upon this foundation, our core innovation lies in proposing Visual-Audio Consistency Group Relative Policy Optimization(VAC--GRPO) equipped with a progressive training strategy. Distinct from standard GRPO, we stratify the training data into four progressive difficulty tiers, guiding the model through a psychologically grounded easy--to--hard cognitive transition. By innovatively coupling this dynamic curriculum scheduler with a multi dimensional, process aware reward mechanism and a reflective learning paradigm, we significantly elevate the model's overall reasoning quality. Extensive experiments on mainstream benchmarks demonstrate that ThinkDeception establishes a new SOTA, significantly outperforming existing methods in both detection accuracy and rationale quality. Ultimately, this work successfully drives the field of deception detection toward interpretable, multimodal cognitive reasoning.