Publications / 2024 Proceedings of the 41st ISARC, Lille, France

From Unstructured Data to Knowledge Graphs: An Application for Compliance Checking Problem

Ankan Karmakar, Chintan Patel, Venkata Santosh Kumar Delhi
Pages 863-871 (2024 Proceedings of the 41st ISARC, Lille, France, ISBN 978-0-6458322-1-1, ISSN 2413-5844)
Abstract:

The rule requirements of a building code are frequently violated to create financially viable designs. These deviations are subjected to condonation by the municipal commissioner if recognizable hardships are faced. The historical concession applications for similar cases are stored in an unstructured manner, creating a barrier to knowledge transfer. The subjective statements given by applicants are composed of logical structure, language, and embedded knowledge that requires years of experience from the domain expert to decipher. A knowledge graph (KG) representation of the problem can capture concepts and represent them visually, which is easy for novice stakeholders to understand. An LLM-based method is used in this study for ontology extraction in the form of concepts and relationships. Also, unstructured input preprocessing and entity disambiguation were performed to evaluate the applicability of KG in this domain. The performance of the proposed method was checked qualitatively in a case study from real-life project examples. The limitations and scopes for improvements were also highlighted. The outcome of this study indicates KG as a potential candidate for knowledge generation from the unstructured archival data of compliance checking. The target audience for this application can be the new architects, reviewers, and programmers working on developing the end-to-end automated compliance checking systems. Finally, applying these AI-based knowledge transfer mechanisms can ignite future research on automated concession applications and approvals, laying a path to the digital transformation of the industry.

Keywords: Code Compliance Checking, Knowledge Graphs, Artificial Intelligence