 This paper proposes a novel approach to solve the problem of decreased accuracy in cross-subject emotion recognition using electroencephalogram, e.g., signals due to negative transfer of data from the source domain. It uses a Frank Coppula model to measure the correlation between the source and target domains, then calculates the maximum mean discrepancy, MMD, to identify the most suitable data for transfer learning. Finally, it employs local tangent space alignment, LTSA, to reduce the dimensionality of the data while preserving its local characteristics. This method was tested on two datasets and showed significant improvements over existing approaches. This article was authored by Yuliang Ma, Wei Qingxiao, Ming Meng, and others.