Methodology for classifying and extracting information with LLM: application on cost estimation case

Cost estimation in building industy largely relies on manually extracting and classifying textual descriptions, a process susceptible to human error. Although recent advancements in Large Language Models (LLMs) hold promise, their application in this domain requires further investigation. This study proposes a methodology to optimize LLM performance validated through the development of a tool that classifies cost descriptions into a three-level hierarchical taxonomy and extracts relevant information organising the data in a database as output. Results demonstrate a F1 score of 0.96 on both tasks contributing to cost estimation automation, reducing manual processing, and enhancing knowledge management within the domain.