딥러닝 자연어처리를 통한 판사의 인지적 과정 추론과 한국 법원 판결 예측 가능성에 관한 연구

초록

Korean judgeʼs unnaturalized cognitive process make it difficult to determine whether similar cases receive judgments; therefore, we aim to understand these processes. This study presents a data analysis method that combines cognitive process knowledge and model techniques. First, the collected data are text that aligns smoothly with the judge's cognitive process, and attempts were made to embed the text into the preprocessed value. In word embedding, FastText was used because it was deemed appropriate for legal analysis. The vector values were learned using Random Forest and Neural network models and enabled judgment to be classified into 81% accuracy. Then, by implementing cognitive maps through language networks and comparing them with the analyzed data, we confirmed that they had the same cognitive process. This study demonstrates the possibility of capturing judgesʼ cognitive process through natural language processing, despite the emergence of challenging legal terminology and lengthy sentences. Additionally, it was possible to deduce the crucial factors and their priorities in judgment through cognitive maps, enabling a comprehensive understanding of factors influencing judgment. From this, it can be concluded that they possess distinct aspect, do share and reflect certain perceptions.

키워드