Emotion Recognition With Knowledge Graph Based on Electrodermal Activity.

Area of Science:

• Affect detection research
• Physiological signal processing
• Knowledge graph applications

Background:

• Electrodermal activity (EDA) sensors are non-invasive tools for measuring skin electrical activity in affect detection.
• Knowledge graphs offer effective data representation, but their integration with physiological signals for emotion recognition in diverse mental states is underexplored.

Purpose of the Study:

• To improve emotion recognition accuracy using EDA signals.
• To investigate the impact of knowledge-related graph features combined with physiological signals on emotion classification.

Main Methods:

• Proposed a deep learning framework for classifying emotional responses from physiological datasets.
• Incorporated gender and age information as embedding feature vectors.
• Extracted time and frequency EDA features and combined them with knowledge embedding feature vectors using a weighted feature fusion method.
• Utilized deep neural networks for optimization.

Main Results:

• The proposed model demonstrated improved performance in valence-arousal emotion recognition.
• Combining Gender-Age Relation Graph (GARG) and statistical feature (SF) vectors enhanced system performance by 4-5% on the PAFEW dataset and 3-2% on the DEAP dataset.

Conclusions:

• The integration of knowledge graph features (GARG) with statistical features (SF) significantly boosts the accuracy of emotion recognition systems based on EDA signals.
• This approach offers a promising direction for more effective and nuanced affect detection.