Traditional dance preservation requires more than video recording because stylistic knowledge is embedded in continuous body motion, rhythm, force, spatial orientation and recurrent movement motifs. This study proposes a genealogy-aware digital reconstruction method for traditional dance movements and conducts comparative analysis across stylistic schools. A corpus containing six traditional dance schools, 144 performance sequences and 3456 segmented movement units is constructed. Each movement unit is encoded by skeletal posture, kinetic variation, rhythm envelope and Laban Movement Analysis descriptors. A multi-relation movement graph is then built by integrating temporal adjacency, morphological similarity, rhythmic affinity, co-occurrence relation and expert correction. The graph is organized into three levels: core motor lexicon, motif family and stylistic branch. Experiments show that the proposed model achieves 0.923 F1, 0.762 ARI and 0.638 style-separation index, outperforming ST-GCN, CTR-GCN, PoseC3D and ablated variants. The reconstructed genealogy reveals that Yangge, Guozhuang and Andai share step-pulse and weight-shift motifs, while Dunhuang dance and Dai Peacock dance are closer in palm arc, torso incline and gaze-turn structures. Error analysis shows that sleeve-sweep movements produce the largest boundary deviation because skeleton data cannot fully capture delayed costume trajectories. The proposed method provides an interpretable computational framework for traditional dance archiving, stylistic comparison and digital transmission.
