Introduction
Patents represent a critical repository of technical and commercial intelligence. However, the exponential growth of patent filings, particularly in China, has made manual expert review increasingly impractical. A significant challenge in automated patent mining is the lack of standardized keywords in many databases and the difficulty in mapping the hierarchical relationships between emerging technologies. This research proposes a systematic framework to transition from raw bibliometric data to a structured topic network, identifying foundational "principal" technologies and their "secondary" extensions.
Methodology: The Four-Stage Framework
The proposed system automates the extraction and mapping of technical topics through four integrated stages, moving from raw data to a relational network.
- Technical Term Extraction: Due to the absence of word delimiters in Chinese text, a hybrid Natural Language Processing (NLP) approach is employed. This involves word segmentation, Part-of-Speech (POS) tagging, and the application of linguistic rules to identify candidate phrases. Final terms are selected based on "termhood" (domain specificity) and "unithood" (phrase stability).
- Topic Detection: Extracted terms are grouped using co-word analysis. An Equivalence Index measures the probability of term co-occurrence, and Coulter’s clustering method is used to form distinct topic clusters.
- Topic Classification: Topics are categorized based on their internal cohesion and external connectivity into three types: Isolated (standalone), Principal (foundational), and Secondary (derivative).
- Network Construction: A directed network is built where nodes represent topic clusters and edges represent external links. The directionality indicates the flow of influence from principal to secondary technologies.
Case Study: Chinese Fuel Cell Patents
The framework was validated using a dataset of 6,346 Chinese fuel cell patents collected from the State Intellectual Property Office (SIPO). From an initial pool of over 28,000 candidate terms, the system refined the set to 1,123 high-quality technical terms after expert evaluation.
Results and Discussion
The analysis identified 62 distinct topics and 10 major sub-domains within the fuel cell field. The resulting network provides a visual roadmap of the technological landscape, distinguishing between core drivers and application-oriented extensions.
A key finding was the identification of "crossroads clusters"—topics that serve as essential joints between different sub-domains. For instance, Topic 24 was found to link five other major topics, suggesting its role as a critical technological enabler. Furthermore, the visualization clearly identifies Topic 1 as the most developed and preferred technology in the domain, acting as a primary driver for various secondary applications.
Conclusion and Future Work
This automated approach effectively maps complex technological relationships without requiring pre-defined taxonomies. While current limitations include sensitivity to manual thresholds and clustering coverage, future work will focus on integrating advanced statistical topic models to enhance the reliability and depth of the network construction, providing better support for competitive intelligence and R&D planning.
Summary updated