<p>Considering the utilization of the correlation between training data, this paper proposes a graph neural network-based deep learning method for the direction of arrival (DOA) estimation by transforming the DOA estimation problem into a node classification problem. Firstly, the covariance matrix of the received signal is used as the network node. Adjacent nodes are divided based on the covariance matrix corresponding to the same incident DOAs while accounting for different signal-to-noise ratios (SNRs) and varying numbers of snapshots. Subsequently, two methods for constructing adjacency matrices are proposed: one based on the correlation of covariance matrix eigenvectors and the other utilizing a feature encoder. These methods enable the construction of a knowledge graph pertinent to DOA estimation. Following this, a deep learning network framework is designed using graph neural networks to enhance the performance of DOA estimation. Finally, the superiority of the proposed method is validated through simulation experiments.</p>