The enduring appeal of old songs across generations raises questions about the factors contributing to their timelessness. This research investigates these factors using a deep learning framework that integrates MusicBERT, a model for symbolic music understanding, with Graph Neural Networks (GNNs). By analyzing a dataset of song features, including lyrics, melody, rhythm, and emotional tone, the study identifies patterns associated with the evergreen status of songs. Key findings emphasize cultural significance, emotional resonance, and universal themes as critical contributors to their lasting popularity. The research offers novel insights into the intersection of music, culture, and technology, with implications for music production, distribution, and cultural heritage preservation.