Motive-based Classification Dataset Research for Similar Case Search Focusing on Criminal on Judgments

With the advancement of artificial intelligence technology, research has been conducted on efficient utilization of large-scale legal documents in the field of legal tech, and services for searching similar cases based on judicial decisions have emerged. In this study, we compare the results of embedding models and investigate criteria for judging similar cases through evaluations by investigative experts, in order to develop an efficient system for searching similar cases targeting criminal lower court judgments. Based on this, we focus on the predominant criterion, which is motivation, among various similarity criteria, and construct a judgment document classification dataset that can assess similarity. We validate its effectiveness using Random Forest, SVM, Decision Tree, KoBERT, and KLUE (bert-base) models.