In recent years, EEG-based sentiment computing has received more and more attention from researchers. Granger causality analysis as a classic feature extraction model has been widely used in sentiment classification models. The model constructs a brain network by calculating the causality relationship between EEG sensors and selects EEG features. Because the original Granger causality analysis uses L2 norm as the loss function and does not perform sparseness. This can cause the results to be affected by noise in the EEG data. Therefore, the researchers put forward Granger causality analysis based on LASSO and Granger causality analysis model based on L1/2 norm to solve the problem of noise. The existing sparse Granger causality analysis model assumes that the connection between each sensor has the same prior probability. However, this article shows that if the correlation of EEG data between each sensor can be added to the Granger causality network as prior knowledge. It can enhance the causal selection ability of the existing sparse Granger causality model, thereby enhancing the model’s EEG feature selection ability. Finally, it can effectively improve the emotion recognition ability of the emotion classifier based on the sparse Granger causality model. In this situation, this paper proposes a new emotional computing model, named Sparse Granger causality analysis model based on sensors correlation(SC-SGA), which uses the correlation between sensors as prior knowledge and combines L1/2-based sparse Granger causality analysis for feature extraction. Finally, the model uses L2 norm logistic regression as the classification algorithm. The experimental results show that the sparse Granger classification model based on sensor similarity proposed in this paper is better than the existing models, and can well identify the emotional state of subjects.

中文翻译：

---

基于传感器相关性的脑电图情绪识别分类稀疏格兰杰因果分析模型

近年来，基于脑电图的情感计算越来越受到研究人员的关注。格兰杰因果分析作为一种经典的特征提取模型，在情感分类模型中得到了广泛的应用。该模型通过计算脑电传感器之间的因果关系并选择脑电特征来构建大脑网络。因为原来的格兰杰因果分析使用L2范数作为损失函数，并没有进行稀疏化。这可能会导致结果受到 EEG 数据中噪声的影响。因此，研究人员提出了基于LASSO的格兰杰因果分析和基于L1/2范数的格兰杰因果分析模型来解决噪声问题。现有的稀疏格兰杰因果分析模型假设每个传感器之间的连接具有相同的先验概率。然而，这篇文章表明，如果每个传感器之间的 EEG 数据的相关性可以作为先验知识添加到 Granger 因果关系网络中。它可以增强现有稀疏格兰杰因果模型的因果选择能力，从而增强模型的脑电特征选择能力。最后，可以有效提高基于稀疏格兰杰因果模型的情感分类器的情感识别能力。针对这种情况，本文提出了一种新的情感计算模型，名为基于传感器相关性的稀疏格兰杰因果分析模型（SC-SGA），它以传感器之间的相关性为先验知识，结合基于 L1/2 的稀疏格兰杰因果分析，特征提取。最后，该模型使用 L2 范数逻辑回归作为分类算法。