A coupling of common–private topological patterns learning approach for cross-subject emotion recognition

• Published in: Biomedical signal processing and control Vol. 105; p. 107550
• Main Authors: Zhang, Haokai, Li, Pengrui, Chang, Hongli, Liu, Shihong, Qin, Yun, Xie, Jiaxin, Wang, Manqing, Gao, Dongrui, Wu, Dingming
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.07.2025
• Subjects: Brain spatial topological patterns; Combinatorial brain patterns (CBPs); CPT-LS; Graph Noise Granularity Adjacency Matrix (GNGAM); Soft Submodality Orthogonalization (SSO); Graph Noise Granularity Adjacency Matrix (GNGAM); Brain spatial topological patterns; Soft Submodality Orthogonalization (SSO); Combinatorial brain patterns (CBPs); CPT-LS
• ISSN: 1746-8094
• DOI: 10.1016/j.bspc.2025.107550

Electroencephalogram (EEG) is a valuable biological signal for emotion recognition due to its high temporal resolution and minimal artifacts. Recent research has focused on understanding brain spatial topological patterns to investigate dynamic mechanisms of brain functional region connectivities and reduce individual variability. This paper introduces a novel learning approach, the Coupling of Common–Private Topological Patterns Learning Strategy (CPT-LS), to address inter-subject differences. The strategy involves encoding and decoding common brain spatial topological patterns, referred to as Combinatorial Brain Patterns (CBPs), shared among subjects in the same emotional state, as well as dynamically encoding and decoding private brain topological patterns (PBPs) unique to each subject in different emotional states. The Graph Noise Granularity Adjacency Matrix (GNGAM) is proposed to enhance the robustness of CBP and PBP learning. At the same time, the soft submodality orthogonalization method (SSO) is introduced to extract pattern-specific information and combine common and private representations to minimize inter-individual differences. The proposed method is evaluated using three public datasets, SEED, SEED-IV, and DEAP, demonstrating superior performance compared to existing approaches and offering a promising direction for advancing affective computing research. •We introduce the CPT-LS to enhance the representation of latent emotional patterns.•We introduce the GNGAM framework to perform adaptive feature-driven updates.•We study the patterns of coupled multi-subject common and private representations.