With the rapid development of social networks and online platforms, the issue of user identity linkage has become a critical research topic in security auditing and recommendation systems. To effectively address the problem of user identity linkage, a method combining multi-layer graph attention networks (GAT) and contrastive learning was proposed. By incorporating the graph attention mechanism, the model was able to capture complex relationships between users more accurately. Additionally, the application of contrastive learning further enhanced the discriminative power and robustness of the node embeddings. A joint optimization strategy was also integrated into the model, enabling the graph attention mechanism and contrastive learning to work synergistically, thereby improving the overall performance of the model. Experimental results demonstrate that the proposed eCGAT-UIL model achieves superior performance across three different datasets, significantly outperforming other existing baseline models.