Scholarlink

<p>Due to the limitations of traditional approaches, lie detection poses significant challenges in real-time scenarios. This study introduces an adaptive multimodal framework that integrates behavioral metrics, textual data, and additional modalities to enhance lie detection accuracy and scalability. The main aim of this research is to determine whether real-time lie detection can be improved using strategic cognitive analysis and a multimodal fusion approach. The proposed framework incorporates the <i>Strategic Interview Technique (SIT)</i> which combines behavioral metrics like response consistency and delay with publicly available datasets, including <i>LIAR</i> and <i>Deceptive Opinion Spam</i>, for textual features. Additional audio-visual data from <i>Bag-of-Lies</i> and <i>DOLOS</i> are incorporated to expand the multimodal framework. Features extracted from each modality are processed using advanced deep learning models, such as Bi-LSTM and CNNs. A novel adaptive fusion layer dynamically weights modalities based on data availability and reliability in real time. The framework is evaluated using accuracy, precision, recall, and F1-score metrics, with a focus on real-time performance. Further the performance was statistically validated using paired t-test. The proposed method achieved a F1-Score of 90.5% and an accuracy of 91.2%, outperforming existing methods. Even with incomplete or noisy input from certain modalities, the real-time adaptability allowed the model to process data efficiently, maintaining robust performance. This study offers significant improvements in accuracy and practicality by demonstrating the potential of an adaptive multimodal approach for real-time lie detection. The findings lay a foundation for scalable deception detection systems applicable in diverse fields such as recruitment, law enforcement and security. Future research could explore incorporating additional data modalities and adaptive question design. However, the proprietary nature of the <i>SIT</i> dataset poses a limitation on public reproducibility and open benchmarking.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive Multimodal Framework for Real-Time Lie Detection Using Strategic Cognitive Analysis

  • 期刊论文

  • 发表日期：

    2025年7月22日

查看全文
  • Debanil Chanda,
  • Rakesh Kumar Mandal

摘要

Due to the limitations of traditional approaches, lie detection poses significant challenges in real-time scenarios. This study introduces an adaptive multimodal framework that integrates behavioral metrics, textual data, and additional modalities to enhance lie detection accuracy and scalability. The main aim of this research is to determine whether real-time lie detection can be improved using strategic cognitive analysis and a multimodal fusion approach. The proposed framework incorporates the Strategic Interview Technique (SIT) which combines behavioral metrics like response consistency and delay with publicly available datasets, including LIAR and Deceptive Opinion Spam, for textual features. Additional audio-visual data from Bag-of-Lies and DOLOS are incorporated to expand the multimodal framework. Features extracted from each modality are processed using advanced deep learning models, such as Bi-LSTM and CNNs. A novel adaptive fusion layer dynamically weights modalities based on data availability and reliability in real time. The framework is evaluated using accuracy, precision, recall, and F1-score metrics, with a focus on real-time performance. Further the performance was statistically validated using paired t-test. The proposed method achieved a F1-Score of 90.5% and an accuracy of 91.2%, outperforming existing methods. Even with incomplete or noisy input from certain modalities, the real-time adaptability allowed the model to process data efficiently, maintaining robust performance. This study offers significant improvements in accuracy and practicality by demonstrating the potential of an adaptive multimodal approach for real-time lie detection. The findings lay a foundation for scalable deception detection systems applicable in diverse fields such as recruitment, law enforcement and security. Future research could explore incorporating additional data modalities and adaptive question design. However, the proprietary nature of the SIT dataset poses a limitation on public reproducibility and open benchmarking.