Scholarlink

<p>Automatic Personality Prediction (APP) is a key area in affective computing, aiming to infer human personality traits from behavioural cues. This study proposes a multimodal deep learning framework for predicting Big Five personality traits using text, audio, and video data. We employ modality-specific and multimodal datasets annotated for the Big Five traits and explore a range of ardy proposes a multimodal deep learning framework for predicting Big Five personality traits using text, audio, and video data. We employ modality-specific and multimodal datasets annotated for the Big Five traits and explore a range of architectures, including transformers, CNNs, and recurrent models (LSTM, CRNN). Results show that audio features, especially MFCC-1 and MFCC-2, offer high predictive power, while sentiment-aware textual embeddings enhance linguistic modelling. Visual features capture non-verbal cues vital for comprehensive trait assessment. We implement both early and late fusion strategies to integrate affective, linguistic, and visual signals, improving robustness and generalisation. To ensure transparency, we incorporate Explainable AI (XAI) techniques, including SHAP and Grad-CAM, to identify influential features across modalities. This enables human-centred analysis and builds trust in model predictions. Our findings highlight the effectiveness of deep multimodal learning for personality modelling and demonstrate how combining behavioural signals with interpretability tools leads to more adaptive and transparent personality-aware AI systems.</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

Personality in 3D: multimodal deep learning framework for big five trait prediction

  • 期刊论文

  • 发表日期：

    2026年3月24日

查看全文
  • Devraj Patel,
  • Sunita V. Dhavale,
  • Bhushan B. Mhetre

摘要

Automatic Personality Prediction (APP) is a key area in affective computing, aiming to infer human personality traits from behavioural cues. This study proposes a multimodal deep learning framework for predicting Big Five personality traits using text, audio, and video data. We employ modality-specific and multimodal datasets annotated for the Big Five traits and explore a range of ardy proposes a multimodal deep learning framework for predicting Big Five personality traits using text, audio, and video data. We employ modality-specific and multimodal datasets annotated for the Big Five traits and explore a range of architectures, including transformers, CNNs, and recurrent models (LSTM, CRNN). Results show that audio features, especially MFCC-1 and MFCC-2, offer high predictive power, while sentiment-aware textual embeddings enhance linguistic modelling. Visual features capture non-verbal cues vital for comprehensive trait assessment. We implement both early and late fusion strategies to integrate affective, linguistic, and visual signals, improving robustness and generalisation. To ensure transparency, we incorporate Explainable AI (XAI) techniques, including SHAP and Grad-CAM, to identify influential features across modalities. This enables human-centred analysis and builds trust in model predictions. Our findings highlight the effectiveness of deep multimodal learning for personality modelling and demonstrate how combining behavioural signals with interpretability tools leads to more adaptive and transparent personality-aware AI systems.