Affective Computing

Affective Computing is an interdisciplinary endeavour that equips machines with the ability to detect, interpret and respond to human emotions. By combining advances in sensor technology, signal processing and machine learning, systems now infer affective states from physiological and behavioural indicators such as facial dynamics, voice prosody, thermal signatures and skin conductance. Approaches range from categorical emotion classification to continuous valence–arousal mapping, often employing deep-learning architectures enhanced by attention and multimodal fusion strategies. Thermal imaging techniques have matured into sensitive, contact-free tools for tracking autonomic responses in naturalistic settings, while convolutional and recurrent neural networks capture subtle muscle movements and temporal patterns. Concurrently, research in human–computer and human–robot interaction emphasises adaptive interfaces that personalise feedback according to moment-to-moment shifts in user affect and long-term user models. Standardisation efforts seek to harmonise data-collection protocols and evaluation metrics, enabling robust deployment of emotionally intelligent technologies across domains such as mental health, education and immersive entertainment.

Research from Nature Portfolio

A classification pipeline has been devised to distinguish emotional versus physical stress by extracting multispectral and tissue oxygenation signals, applying wavelet-based frequency analysis and encoding features via a multi-neighbor vector model, then classifying with a long short-term memory network that achieves over 90 per cent accuracy. Another study expands the emotional taxonomy to ten discrete states using a large, richly annotated dataset and explores both transfer-learning with Inception-V3 and MobileNet-V2 backbones and bespoke CNNs trained from scratch, reporting peak test accuracies near 96 per cent and F1-scores of 0.95. A further investigation integrates squeeze-and-excitation modules with residual blocks and self-attention distillation to refine local facial feature representations, demonstrating that nasal and perioral regions reliably drive expression-recognition performance in the wild.

Research from all publishers

A nonlinear cross-mapping study of facial thermography under rest and Stroop-induced stress quantifies strong coupling between nasal temperature and electrodermal activity, revealing gender-specific patterns in autonomic regulation. A comprehensive review of thermal infrared imaging in social robotics highlights how involuntary cutaneous cues complement visual expression analysis, enabling robots to detect user stress, engagement and frustration in real time. In immersive virtual-reality experiments, researchers demonstrate that mean valence and arousal trajectories across an entire episode outperform peak-and-end measures for predicting retrospective experience, suggesting affective evaluation systems should aggregate full-sequence data rather than focus solely on extremal moments.

Affective Computing publication trend

The graph below shows the total number of articles in affective computing across all publications each year (not limited to Nature Index journals).

Technical terms

Valence: The pleasantness–unpleasantness dimension of an emotional state.

Arousal: The intensity or activation level component of affect.

Multimodal fusion: The integration of diverse data sources (e.g., thermal, visual, physiological) to enhance emotion inference.

Squeeze-and-excitation network: A neural module that adaptively recalibrates channel-wise feature responses using global context.

Cross-mapping analysis: A nonlinear time-series technique for assessing bidirectional coupling strength between physiological signals.

References

1. New Frontiers for Applications of Thermal Infrared Imaging Devices: Computational Psychopshysiology in the Neurosciences. Sensors (2017).
2. Estimation of continuous valence and arousal levels from faces in naturalistic conditions. Nature Machine Intelligence (2021).
3. Affective Computing Needs Personalization—And a Character?.
4. Facial thermal imaging: A systematic review with guidelines and measurement uncertainty estimation. Measurement (2025).
5. Classification of emotional stress and physical stress using a multispectral based deep feature extraction model. Scientific Reports (2023).
6. Autonomic Regulation of Facial Temperature during Stress: A Cross-Mapping Analysis. Sensors (2023).
7. Thermal Infrared Imaging-Based Affective Computing and Its Application to Facilitate Human Robot Interaction: A Review. Applied Sciences (2020).
8. From Experience to Memory: On the Robustness of the Peak-and-End-Rule for Complex, Heterogeneous Experiences. Frontiers in Psychology (2019).