Graph representation learning via collaborative online adversarial knowledge transfer for EEG based emotion recognition: application to hearing-impaired subjects

Electroencephalogram (EEG)‐based emotion recognition enables objective mental‐health monitoring and affective brain–computer interfaces (BCIs). However, hearing‐loss‐related neural reorganization causes strong inter‐individual EEG heterogeneity; intrinsic EEG non‐stationarity, inter‐subject distribution shifts, static graph topology limitations and pre‐trained‐teacher‐based distillation further degrade cross‐subject recognition performance. To tackle these challenges, this paper proposes a Spearman-Band Attention Graph Network with Online Adversarial Distillation. We construct a multi-view feature enhancement module integrating frequency band attention, convolutional block attention module (CBAM) channel-spatial joint attention and Spearman correlation operators to extract EEG features from three dimensions: frequency bands, electrode space and global inter-electrode feature association. A GATv2-based graph encoder with residual skip connections is utilized to alleviate the over-smoothing issue of deep graph networks and capture multi-scale topological relationships among electrodes. Furthermore, a two-layer collaborative online distillation framework is devised: global distillation based on KL divergence aligns prediction distributions of multiple student models, while local cyclic adversarial distillation oriented to node embeddings mines latent graph structural features. A diversity constraint is introduced to prevent model collapse. Validations were conducted on the self-built HIED dataset of individuals with hearing impairment and the public SEED dataset. The subject-dependent accuracies reached 80.54% and 96.65%, respectively, and the leave-one-subject-out cross-subject accuracies reached 70.56% and 86.67%, respectively. Visualizations show more dispersed parieto-occipital model-derived spatial-attention and feature-association patterns in individuals with hearing impairment.

Authors

Institutions

Publication Details

Journal
Measurement Science and Technology
Published
2026-11-18
DOI
https://doi.org/10.1088/1361-6501/aeb711
Primary Topic
Emotion and Mood Recognition
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Graph representation learning via collaborative online adversarial knowledge transfer for EEG based emotion recognition: application to hearing-impaired subjects

Yu Song, Qing Wang, Zhonghao Zhou, Nan Geng et al.
Measurement Science and Technology
Emotion and Mood Recognition
article

Graph representation learning via collaborative online adversarial knowledge transfer for EEG based emotion recognition: application to hearing-impaired subjects

Yu Song, Qing Wang, Zhonghao Zhou, Nan Geng, Yushuo Jia, Jinnuo Wang, Weijing Han
article en

Abstract

Electroencephalogram (EEG)‐based emotion recognition enables objective mental‐health monitoring and affective brain–computer interfaces (BCIs). However, hearing‐loss‐related neural reorganization causes strong inter‐individual EEG heterogeneity; intrinsic EEG non‐stationarity, inter‐subject distribution shifts, static graph topology limitations and pre‐trained‐teacher‐based distillation further degrade cross‐subject recognition performance. To tackle these challenges, this paper proposes a Spearman-Band Attention Graph Network with Online Adversarial Distillation. We construct a multi-view feature enhancement module integrating frequency band attention, convolutional block attention module (CBAM) channel-spatial joint attention and Spearman correlation operators to extract EEG features from three dimensions: frequency bands, electrode space and global inter-electrode feature association. A GATv2-based graph encoder with residual skip connections is utilized to alleviate the over-smoothing issue of deep graph networks and capture multi-scale topological relationships among electrodes. Furthermore, a two-layer collaborative online distillation framework is devised: global distillation based on KL divergence aligns prediction distributions of multiple student models, while local cyclic adversarial distillation oriented to node embeddings mines latent graph structural features. A diversity constraint is introduced to prevent model collapse. Validations were conducted on the self-built HIED dataset of individuals with hearing impairment and the public SEED dataset. The subject-dependent accuracies reached 80.54% and 96.65%, respectively, and the leave-one-subject-out cross-subject accuracies reached 70.56% and 86.67%, respectively. Visualizations show more dispersed parieto-occipital model-derived spatial-attention and feature-association patterns in individuals with hearing impairment.

Measurement Science and TechnologyVol. 37(46)
Tianjin University of Technology (CN)
National Natural Science Foundation of China
Good health and well-being
Emotion and Mood Recognition
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon