Measuring Text-Based Task Similarity to Enable Personalized Learning in Modular Construction: A Case Study in HVAC Product Manufacturing

In workforce training for manufactured construction, managers have relied on their tacit knowledge to manage workers’ learning curves, resulting in redundancies and less effective skill acquisition. While previous studies demonstrate that skill acquisition is more effective when tasks align with prior knowledge, less attention has been given to identifying similarities across different workflows. This study proposes a framework for enhancing skill acquisition by measuring task similarity between workflows. Using operation manuals, we represented a knowledge graph and vectorized each item to calculate similarity using a large language model (LLM). Validation was conducted using standard operating procedures (SOPs) from two different heating, ventilation, and air conditioning (HVAC) product production lines. This study demonstrates that items with specific roles exhibit higher similarity than those with general roles and highlights the potential to identify the most similar task in different workflows. The findings can guide research on identifying tasks requiring focused training.