Answers are grounded in this paper and cite their sources.
Emotion recognition from electroencephalogram (EEG) signals has emerged as a key research direction in affective computing and brain–computer interface (BCI) systems due to its objectivity and rich temporal dynamics. In this work, we propose a novel Spiking Graph-LSTM framework integrated with multi-head temporal attention for accurate EEG-based emotion recognition. The proposed model jointly captures temporal dependencies and spatial relationships among EEG channels by unifying biologically inspired spiking neural encoding with graph convolutional learning. Initially, raw EEG signals are denoised using band-specific filtering and wavelet decomposition, then segmented into emotion-related epochs. Discriminative features are extracted from the time, frequency, and power spectral domains, and high-dimensional representations are efficiently transformed into spike trains through rate-based encoding. A multi-head temporal attention mechanism then adaptively emphasizes salient EEG segments by learning their relative importance across time. In parallel, a Graph Convolutional Network augmented with LSTM units dynamically models inter-feature correlations using a learned adjacency matrix, enabling practical propagation of spatial and temporal information. The integration of residual segment connections and attention-weighted embeddings results in highly expressive emotion representations. Experimental results demonstrate clear separability between positive and negative emotional states, as visualized using t-SNE. The proposed approach achieves 99% accuracy, significantly outperforming both conventional GCN-based and attention-only models, highlighting its robustness and effectiveness for EEG-based emotion recognition.
No takes yet. Share an insight, caveat, or question.
Gudikandula et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: