Numerical Element Calculation Method for Criminal Cases with Case Knowledge

The calculation of numerical elements in criminal cases aims to determine the final values related to the suspect from the case description, providing the necessary support for the sentence prediction. Although models based on graph neural networks have numerical computing capabilities, existing composition methods cannot be well applied to criminal cases. This paper proposes a numerical element calculation method for criminal cases that integrates case knowledge. Firstly, we construct a heterogeneous graph of “case knowledge” for the description of criminal cases. Then we apply the graph attention network inference on the heterogeneous graph of "case knowledge" to fully capture the semantic information between suspects and numerical values. Finally, an arithmetic symbol is assigned to the numerical value to obtain an arithmetic expression for calculation. The experimental results on the multi cause criminal cases from a certain provincial court and the 2021 LAIC dataset show that the proposed method improves the average accuracy by 3% compared to the baseline model.