Evaluation System for Reasoning Description of Judgment Documents Based on TensorFlow CNN

Abstract

In order to improve the quality of the judgment documents, the state and government have introduced laws and regulations. However, the current status of trials in our country is that the number of cases is very large. Using system to verify the documents can reduce the burden on the judges and ensure the accuracy of the judgment. This paper describes an evaluation system for reasoning description of judgment documents. The main evaluation steps include: Segmenting the front and back of the law; extracting the key information in the document by using XML parsing technology; constructing the legal exclusive stop word library and preprocessing inputting text; entering the text input into the model to get the text matching result; using the “match keyword, compare sentencing degree” idea to judge whether the logic is consistent if it is the evaluation of “law and conclusion”; integrating the calculation results of each evaluation subject and feeding clear and concise results back to the system user. Simulation of real application scenarios was conducted to test whether the reasoning lacks key links or is insufficient or the judgment result is unreasonable. The result show that evaluation speed of each document is relatively fast and the accuracy of the evaluation of the common nine criminal cases is high.