특허문서의 IPC 계층별 분류기 생성: B섹션 기준 섹션부터 서브클래스까지 = Generation of Hierarchical Classifiers for Patent Documents Based on IPC Levels: From Section to Subclass Focusing on Section B

다국어 초록 (Multilingual Abstract)

Since the establishment of the Korean Intellectual Property Office (KIPO), patent applications across various fields have continued to increase in Korea, and this upward trend is expected to persist. However, the patent examination process still takes more than five years. To address this issue, this paper proposes a classifier to automate IPC code prediction within the patent examination workflow, aiming to reduce examination time and improve efficiency. Given that Section B contains a large number of classes but relatively few patents compared to other sections, the scope of the proposed classifier is limited to the subclasses of Section B. Furthermore, to evaluate the impact of independent claims on classification performance, we constructed datasets both with and without independent claims. We then developed classifiers using four different techniques—Multinomial Naïve Bayes, XGBoost, LightGBM, and Random Forest— and compared their performance.