Developing Automatic-Labeled Topic Modeling Based on SAO Structure for Technology Analysis.

Area of Science:

• Information Science
• Technology Management
• Computational Linguistics

Background:

• Topic modeling is crucial for identifying technology trends and forecasting advancements.
• Existing auto-labeling methods in patent documents have limitations in quantitative evaluation and capturing technological hierarchy.
• Traditional approaches often overlook the functional significance of technologies.

Purpose of the Study:

• To introduce a subject-action-object (SAO) based approach for topic modeling in patent documents.
• To overcome limitations of existing auto-labeling methodologies by utilizing the "Bag of SAO" concept.
• To enhance the accuracy and interpretability of technology trend analysis.

Main Methods:

• Developed an auto-labeling methodology combining SAO-based topic modeling and scoring.
• Integrated text summarization and network analysis for comprehensive evaluation.
• Incorporated a hierarchical structure based on Cooperative Patent Classification (CPC) subclasses for enhanced interpretability.

Main Results:

• The proposed model demonstrated effectiveness in capturing functional meanings within the technological context, evaluated using ROUGE, relevance, coverage, and discrimination scores.
• The SAO-based approach improved the accuracy of topic labels and provided deeper semantic insights.
• The hierarchical structure enhanced the view of technological development and trends.

Conclusions:

• The SAO-based auto-labeling methodology offers a more accurate and interpretable approach to topic modeling in patent documents.
• This approach contributes to more efficient technology management, innovation, and policy-making.
• The study lays the groundwork for refined methodologies in technology forecasting and diagnosis.