Related Experiment Video
Updated: Oct 5, 2025
05:51
Exploring the Use of Isolated Expressions and Film Clips to Evaluate Emotion Recognition by People with Traumatic Brain Injury
Published on: May 15, 2016
9.2K
Investigating EEG-based functional connectivity patterns for multimodal emotion recognition
Xun Wu1, Wei-Long Zheng1,2, Ziyi Li1
1Center for Brain-Like Computing and Machine Intelligence, Department of Computer Science and Engineering, Shanghai Jiao Tong University, 800 Dongchuan Rd., Shanghai 200240, People's Republic of China.
Journal of Neural Engineering
|January 31, 2022
Summary
This study introduces a new method for emotion recognition using electroencephalography (EEG) functional connectivity networks. The novel approach significantly improves accuracy by analyzing brain network features, outperforming single-channel methods.
Area of Science:
- Neuroscience
- Machine Learning
- Affective Computing
Background:
- Existing emotion recognition from electroencephalography (EEG) often overlooks the functional connectivity between brain regions.
- Single-channel feature extraction methods limit the comprehensive understanding of brain activity during emotional states.
Purpose of the Study:
- To propose a novel algorithm for selecting emotion-relevant critical subnetworks from EEG functional connectivity networks.
- To investigate the efficacy of EEG functional connectivity network features (strength, clustering coefficient, eigenvector centrality) for emotion recognition.
- To develop a multimodal emotion recognition model integrating EEG connectivity features with eye movement data.
Main Methods:
- Constructed brain networks using correlations between pairs of EEG signals.
- Developed a critical subnetwork selection algorithm by averaging brain network matrices for specific emotion labels.
- Employed deep canonical correlation analysis to integrate EEG connectivity features and eye movement data in a multimodal model.
- Evaluated the model on SEED, SEED-V, and DEAP public datasets.
Main Results:
- The proposed strength feature demonstrated superior performance compared to state-of-the-art single-channel features.
- Achieved high classification accuracies: 95.08±6.42% (SEED), 84.51±5.11% (SEED-V), and 85.34±2.90% (arousal) / 86.61±3.76% (valence) (DEAP).
- Networks constructed with 18 channels yielded performance comparable to 62-channel networks, simplifying practical application.
Conclusions:
- The proposed EEG functional connectivity networks combined with an emotion-relevant critical subnetwork selection algorithm offer a successful approach to emotion recognition.
- This method effectively captures inter-channel information, advancing the field of affective computing.
- The findings suggest a more efficient and effective method for emotion recognition using reduced channel setups.
