脑电情绪识别的CiteSpace知识图谱分析

CiteSpace Knowledge Graph Analysis of EEG-Based Emotion Recognition

摘要: 为探究国内脑电情绪识别领域的演变路径与前沿热点，本研究采用文献计量学方法，选取2011~2024年间CNKI与Scopus数据库收录的155篇高被引核心文献为精选样本，利用CiteSpace软件构建知识图谱，从发文趋势、科研机构、核心作者及关键词共现等维度开展定量分析。结果显示，该领域历经萌芽与稳步增长后，于2021年步入爆发式发展阶段，杭州电子科技大学及孔万增等学者在该领域具有显著影响力。研究协作网络呈现典型的“大分散、小聚类”特征，CNKI与Scopus网络密度分别仅为0.0214与0.0148，跨机构深度协同仍受壁垒限制。技术逻辑已实现从“特征工程驱动”向“数据驱动与脑机理启发式建模”的范式跨越：特征表征由功率谱密度向非线性微分熵及脑网络功能连接深化；算法模型由传统映射转向卷积神经网络、图神经网络及Transformer混合架构。多模态融合、迁移学习及非欧空间拓扑建模已成为当前国际竞争的前沿方向。未来研究应聚焦于构建契合大脑生物特性的启发式计算模型，并致力于复杂环境下鲁棒特征的提取及其在临床医疗、人机交互场景的落地转化。

Abstract: To explore the evolutionary trajectory and research frontiers of EEG-based emotion recognition in China, this study employs bibliometrics methods to analyze 155 high-impact documents indexed in CNKI and Scopus from 2011 to 2024. Utilizing CiteSpace for knowledge graph construction, quantitative and qualitative analyses were conducted across multiple dimensions, including publication trends, institutional contributions, core authors and keyword co-occurrence. Findings indicate that after a steady growth phase, the field entered explosive development in 2021, with Hangzhou Dianzi University and scholars such as Kong Wanzeng exerting significant influence. The collaboration network reveals a “broadly dispersed, locally clustered” pattern, with network densities of 0.0214 (CNKI) and 0.0148 (Scopus), highlighting persistent structural barriers to cross-institutional synergy. The technical paradigm has shifted from “feature-engineering-driven” to “data-driven and brain-inspired modeling”. Feature representation has advanced from Power Spectral Density toward non-linear Differential Entropy and functional connectivity, while algorithmic architectures have transitioned from traditional models to hybrid CNN, GNN, and Transformer frameworks. Currently, multi-modal fusion, transfer learning, and non-Euclidean topological modeling represent the leading edges of international competition. Future research should prioritize constructing heuristic computational models aligned with the biological characteristics of the brain, improving feature robustness in complex environments, and facilitating practical implementation in clinical medicine and human-computer interaction.