Evaluating the Rationality of Judicial Decision with LSTM-Based Case Modeling

In the judicial domain, enormous growth of various legal data has greatly increased the workload of law professionals, which may reduce the efficiency of case handling or even threaten the judicial justice. In this paper, we present an approach to the evaluation of judgment rationality with thorough analysis of Chinese judicial decision texts. A novel metric, judgment deviation, is proposed to measure the likelihood of a certain case's misjudgment according to the judgment results of its similar cases. We employ long short-term memory (LSTM) model to implement case modeling by extracting the feature vector based on the key factors that affect the judicial decision. Experiments on a real dataset including about 25,000 judgments of theft cases and artificial validation demonstrate that the results of the proposed evaluation approach conform to reality.