Advancing Domain-Specific Knowledge Graph Construction: Influencing Innovations in Customized Knowledge Graph Technologies

Impact summary

This research, conducted in collaboration with Rolls-Royce at NTU Corporate Lab, resulted in the development of HDSKG, a novel method for harvesting domain-specific knowledge graphs from web content. The research has significantly impacted the field of knowledge graph construction, directly influencing Oracle International Corporation's patented technology, titled "TECHNIQUES FOR BUILDING A KNOWLEDGE GRAPH IN LIMITED KNOWLEDGE DOMAINS" (WO2020037217A1, 11625620US and EP3635650A1), the application filed in 2020 and granted in 2023. The HDSKG method has received 77 citations, reflecting its influence and recognition in the academic and professional communities.

Research and engagement activities leading to impact

Rationale: Knowledge graphs play a crucial role in various applications such as search result ranking, recommendation systems, and exploratory search. Constructing domain-specific knowledge graphs poses significant challenges due to the complexity and diversity of natural language across different domains. This research aimed to overcome these challenges by developing an efficient, automated method for extracting domain-specific concepts and relations from web content.

Research Work: The project resulted in the creation of HDSKG, a method that integrates dependency parsing with rule-based techniques to automatically extract domain-specific relation triples (subject, verb phrase, object) from webpages. The research was validated using data from Stack Overflow, a software engineering Q&A platform, leading to the construction of a knowledge graph with 35,279 relation triples, 44,800 concepts, and 9,660 unique verb phrases.

Engagement Activities: The research was conducted in collaboration with Rolls-Royce researchers at NTU Corporate Lab, utilizing their expertise and resources to refine the method and assess its performance. The end-users involved in this research included academic researchers and industry professionals specializing in data science, artificial intelligence, and knowledge management.

Relevant Inputs: The project benefited from access to the Stack Overflow dataset, computational resources provided by NTU Corporate Lab, and domain expertise from Rolls-Royce researchers.

Research outputs associated with the impact

The key research output was the paper titled "HDSKG: Harvesting Domain-Specific Knowledge Graph from Content of Webpages," presented at the 24th IEEE International Conference on Software Analysis, Evolution, and Reengineering (SANER 2017), a CORE-ranked A conference. This paper has been cited as one of 19 scholarly outputs in Oracle International Corporation’s patent on techniques for building knowledge graphs in limited knowledge domains. The HDSKG method has garnered 77 citations, according to Google Scholar, highlighting its impact and relevance.

Researcher involvement

As a co-author and collaborator in this project, I played a crucial role in the conceptualization, development, and validation of the HDSKG method. My responsibilities included co-developing the machine learning algorithm for estimating domain relevance, integrating the dependency parser with rule-based methods, and evaluating the system's performance. My active involvement in this project, in collaboration with Rolls-Royce researchers, ensured that the research addressed practical challenges in domain-specific knowledge graph construction.

Outcomes of research leading to impact

The HDSKG method has been implemented and adopted by Oracle International Corporation, as demonstrated by their citation of our work in the patent (11625620US) granted in 2023. Oracle's patent builds on the concepts introduced in our research, particularly in rule-based extraction and machine learning for domain-specific applications, showcasing the practical application and adoption of our research in the development of innovative technologies.

Beneficiaries of the impact

Reach: The primary beneficiaries of this research include technology companies like Oracle, particularly those involved in data science and artificial intelligence. Additionally, academic researchers and professionals in knowledge management and natural language processing have benefited from the advancements made in knowledge graph technology.

Relationships: The collaboration with Rolls-Royce at NTU Corporate Lab and the subsequent adoption of our research by Oracle highlight the strong industry-academic partnerships that facilitated the impact of this research.

Details of the impact achieved

Significance: The impact of this research is substantial, as it has directly influenced Oracle’s patented technology for constructing customized knowledge graphs. These knowledge graphs are critical for improving the accuracy and efficiency of data-driven applications across various industries.

Evidence: The influence of our research is evidenced by Oracle International Corporation's citation of our paper in their patent families (WO2020037217A1, 11625620US and EP3635650A1), which acknowledges the contribution of the HDSKG method to their knowledge graph construction techniques in limited knowledge domains. Additionally, the 77 citations received by the HDSKG paper according to Google Scholar further demonstrate the broader impact and significance of this work within the academic and professional communities.

Impact date	2020
Category of impact	Other Impact
Impact level	International