To address challenges in tax services and achieve digital transformation in tax administration, this paper proposes a tax digital management system and explores its application strategy within large corporate groups. This paper outlines the research feasibility and primary design requirements of the system, providing a foundation for subsequent development of the tax digital management system. First, a classification coding model for tax invoice transaction behaviors is proposed. This model integrates an attention mechanism to extract multi-level features from short text within tax invoice behavior data. Second, a high-concurrency transaction behavior coding query system based on this model is constructed to support diverse query requests. Third, Baidu Maps API is utilized to reveal geographical distribution disparities within the tax industry. Finally, a knowledge graph visualization of taxpayers is presented based on multi-source tax data. Results demonstrate that the proposed model achieves an average accuracy rate of 98% in transaction behavior recognition. This system enhances the precision of tax planning and improves the stability of tax management execution, offering significant reference value for modernizing tax management in large conglomerates.
