자연어처리와 기계학습을 활용한 기술 특허 분류

With the advent of the big data era, machine learning models, including artificial neural networks, have had a wide-ranging impact on various fields such as medicine, genomics research, and corporate management. Despite this, domestic research in legal tech, particularly applying natural language processing and machine learning to technical patent analysis, has not sufficiently developed. This study designs a system for classifying patents on Carbon Dioxide Capture and Utilization (CCU) based on patent data, natural language pre-processing techniques, and machine learning models, and compares and analyzes accuracy, kappa coefficient, and F1-score. The main findings are summarized as follows: First, in classifying five types of CCU technologies, the performance was observed in the order of gradient boosting, random forest, and decision trees. This confirms that random forest and gradient boosting models, which apply bagging and boosting techniques, respectively, provide superior learning performance over single decision trees. Second, similar performance was observed in classifying technologies based on the abstract and first claim of patents. This suggests that the extraction of important keywords as nouns during the natural language processing is a significant factor. This research is meaningful as it applies natural language pre-processing and machine learning models to the classification of CCU technology patents for the first time, presenting the potential for applying robotic automation technology to automate repetitive tasks.