Unsupervised Approach for Specialized Vocabulary Creation and Enrichment: A Case Study in the Multidisciplinary Building Sector

The exponential growth of digital information has exposed organizations to unprecedented challenges in managing and structuring their knowledge repositories. In the context of knowledge management, the ability to extract, organize, and use relevant information from large collections of documents has become a critical factor for operational efficiency and informed decision-making. However, identifying necessary knowledge sources and building appropriate knowledge bases represents a significant and time-consuming barrier. In this article, we address these challenges by leveraging advanced Natural Language Processing (NLP) techniques, particularly in combination with Large Language Models (LLMs), to facilitate the selection of more representative keywords for the creation and enrichment of vocabularies for knowledge management purposes. We explore the application of clustering techniques combined with NLP-driven keyword extraction to support the construction of specialized vocabularies that address the multidisciplinary nature of the content at CSTB, a French scientific research center focused on building science. We applied a pipeline with two approaches for keyword extraction: document-based clustering and chunk-based clustering. We provide a detailed overview of the proposed pipeline, present the results of our experiments, and describe the human validation process used to evaluate these results.