자연어 처리 기법을 활용한 판례 기반 모욕성 문장 분석 시스템

다국어 초록 (Multilingual Abstract)

With the advent of social media, there has been a significant rise in the number of individuals openly engaging in online insults, thereby emerging as a notable social issue. In order to protect users from unpleasant experiences, online portal sites employ database-driven profanity filters to render offensive content invisible. However, these filters often fall short in preventing instances where individuals insult others by judging certain words, not classified as profanity, as offensive within a specific context. Therefore, this paper aims to propose a precedent-based insult sentence analysis system, utilizing advanced natural language processing techniques. The system leverages a deep learning model, rooted in precedents, to infer the likelihood of guilt associated with an insult sentence. Furthermore, the system presents users with comparable precedents through similarity analysis. The ultimate model was chosen based on the accuracy of the test dataset and the training dataset while ensuring enhanced accessibility by deploying it on the web.