Automated Development of Domain-specific Thesaurus for Text Mining in the Construction Industry

Text mining techniques are actively utilized in the construction industry to automatically analyze documents and extract experiential knowledge from the field.However, challenges persist in leveraging Natural Language Processing (NLP) and Large Language Models (LLMs) due to the specialized terminology and unique expressions prevalent in the construction domain.This study proposes a methodology for the automated construction of a domain-specific thesaurus tailored to text mining applications in the construction industry.Using construction specification data, we employed co-occurrence analysis, Word2Vec, and cosine similarity to construct a thesaurus comprising 748,890 co-occurring words and 172,650 word embeddings.Furthermore, word associations and their interrelationships were analyzed and visualized using a word network.The thesaurus developed in this study facilitates precise correction of word recognition errors by leveraging domain-specific characteristics, thereby significantly enhancing the accuracy of text analysis for construction-related documents.