Existing methods for detecting music plagiarism rely on diverse criteria but remain inherently subjective. Altered audio files (e.g., speed adjustments, pitch changes, etc.) are difficult to accurately determine infringement using traditional similarity measures. The ability to detect file modifications, alongside plagiarized content, is critical in protecting copyright in complex musical environments and holds considerable promise for real-world applications. This study utilizes Graph Attention Network (GAT) framework to improve the precision of similarity assessments between songs. Experimental evaluations demonstrate that the proposed method achieves a 7% improvement in detecting plagiarized and altered audio files compared to conventional non-network-based models. These findings underscore the efficacy of leveraging graph edge attention to enhance acoustic similarity analysis within the network.