Park Minyoung, Kim Sunhye, Yoon Byungun
Master student, Department of Industrial & Systems Engineering, School of Engineering, Dongguk University, Seoul, Korea.
Post-doctoral researcher, Department of Industrial & Systems Engineering, School of Engineering, Dongguk University, Seoul, Korea.
PLoS One. 2025 Aug 26;20(8):e0330275. doi: 10.1371/journal.pone.0330275. eCollection 2025.
Topic modeling has become essential for identifying emerging technology trends, detecting technological concepts, and forecasting advancements. This study introduces a subject-action-object (SAO) based approach to overcome the limitations of existing auto-labeling methodologies in patent documents. In particular, by utilizing the "Bag of SAO" concept, the study aims to construct topic modeling itself on an SAO basis, thereby clarifying the complex relationships within technology. Traditional auto-labeling methods often lack sufficient quantitative evaluation metrics and overlook the functional significance and hierarchical structure of technologies. To address these challenges, we propose an auto-labeling methodology that combines SAO-based topic modeling and scoring with text summarization and network analysis. The proposed model's effectiveness was evaluated using the ROUGE score alongside others such as relevance, coverage, and discrimination, showing its ability to capture functional meanings within the technological context. To enhance interpretability, we integrated a hierarchical structure based on CPC subclasses, offering a more comprehensive view of technological development and trends. This approach is expected to improve the accuracy of topic labels while providing deeper semantic insights, contributing to more efficient technology management. This study illustrates how SAO-based auto-labeling methodologies can be applied in the field of technology management, highlighting their potential applications in technology innovation, policy-making, and industry applications. Furthermore, by integrating the SAO structure, this research is anticipated to lay the groundwork for developing more refined methodologies for technology forecasting and diagnosis in future studies. Through this, we hope to gain a clearer understanding of the directions of technological advancement and provide strategic insights for the development of new technologies.
主题建模已成为识别新兴技术趋势、检测技术概念和预测技术进步的关键。本研究引入了一种基于主谓宾（SAO）的方法，以克服专利文献中现有自动标注方法的局限性。具体而言，通过利用“SAO包”概念，该研究旨在基于SAO构建主题建模本身，从而厘清技术内部的复杂关系。传统的自动标注方法往往缺乏足够的定量评估指标，并且忽视了技术的功能重要性和层次结构。为应对这些挑战，我们提出了一种将基于SAO的主题建模和评分与文本摘要及网络分析相结合的自动标注方法。使用ROUGE分数以及相关性、覆盖率和区分度等其他指标对所提出模型的有效性进行了评估，结果表明其能够在技术背景下捕捉功能含义。为了增强可解释性，我们整合了基于美国专利分类号（CPC）子类的层次结构，从而更全面地呈现技术发展和趋势。这种方法有望提高主题标签的准确性，同时提供更深入的语义洞察，有助于实现更高效的技术管理。本研究阐述了基于SAO的自动标注方法如何应用于技术管理领域，突出了它们在技术创新、政策制定和行业应用中的潜在应用。此外，通过整合SAO结构，预计本研究将为未来研究开发更精细的技术预测和诊断方法奠定基础。通过这样做，我们希望更清楚地了解技术进步的方向，并为新技术的开发提供战略见解。
Zotero 插件
