Multi-Task Learning-Based AI System for Legal Judgment Logic Prediction in Economic Law Litigation

Abstract

This study proposes a judgment logic prediction system tailored for economic law litigation by leveraging a multi-task deep learning architecture. The system integrates natural language processing and structured data analysis to address the complexity and volume of legal cases. A dataset of 1,000 corporate bankruptcy cases was constructed, with 800 for training and 200 for testing. Experimental results demonstrate that the proposed MTLPN model achieves an accuracy of 89.3%, outperforming traditional models such as SVM (79.2%) and Random Forest (81.7%). The mean average precision reached 81.4%, and the system reduced average reasoning time to 1.2 seconds per case. Additionally, judgment consistency among judges increased from 78.2% to 91.3% in test scenarios. This study highlights significant improvements in efficiency, accuracy, and transparency in economic law case handling, validating the system’s value in supporting judicial decision-making and promoting fairer outcomes.