Research and Application of Anti-Money Laundering Transaction Detection based on DeepSeek-R1 Small Model using Knowledge Distillation

Anti-money laundering transaction identification ensures that law laws are followed, allowing for a safe financial environment free of illegal activity, money laundering and terrorist financing. In this paper, DeepSeek-R1 small model-based knowledge distillation technology is used to propose a new anti-money laundering transaction identification method. We present a layered hybrid architecture based on the well-known anti-money laundering structure, where we enhance the performance of small models achieving better performance compared to the traditional approaches without affecting performance. In general, the proposed model performs knowledge distillation from a complicated large-scale teacher model to a simple student model in a teacher-student framework. This knowledge distillation technology allows the small model to have high detection accuracy, but also enjoys faster inference speed and lower computational overhead. Experimental results reveal that the proposed model outperforms the classic method significantly in the detection accuracy and processing efficiency, and it also provides an effective solution for the real-time anti-money laundering system.