Twisted bilayer graphene (TBG) has drawn considerable attention due to its angle-dependent electrical, optical, and mechanical properties, yet preparing and identifying samples at specific angles on a large scale remains challenging and labor-intensive. Here, a data-driven strategy that leverages Raman spectroscopy is proposed in combination with deep learning to rapidly and non-destructively decode and predict the twist angle of TBG across the full angular range. By processing high-dimensional Raman data, the deep learning model extracts hidden information to achieve precise twist angle identification. This approach is further extended to a 2D plane, enabling accurate orientational mapping within individual samples. Through interpretability analysis, the model is validated in conjunction with first-principles theoretical calculations, ensuring robust and explainable results. This data-driven methodology not only facilitates efficient TBG characterization but also introduces a broadly applicable framework for studying other angle-dependent 2D materials, thereby advancing the field of material spectroscopy and analysis.