Development of Multimodal Recommender System for Advancing Design for Safety with Expert Validation

Construction projects often struggle to integrate safety considerations effectively and identify hazards comprehensively during the design phase. Design for Safety (DfS) is a promising approach to embedding safety into the design phase, but traditional DfS practices are often subjective, time-consuming, and limited by human error. These challenges result in inconsistent safety outcomes and hinder the adoption of DfS in real-world projects. This study addresses these limitations by developing a novel multimodal recommender system that combines textual and image data to provide tailored DfS recommendations. By employing cosine similarity for feature analysis, the system integrates user input to generate relevant and actionable safety suggestions. User interaction is facilitated through a user-friendly web application, which enables construction professionals to access and apply recommendations efficiently. The system was evaluated by industry experts, achieving an average normalized discounted cumulative gain (NDCG) score of 0.63, demonstrating its strong alignment with expert relevance assessments. Additionally, the system revealed notable strengths, including ease of use (3.37/5), intuitiveness (3.40/5), and relevance of recommendations (3.73/5). Correlation analysis identified strong links between these factors and overall user satisfaction. This study introduces a cutting-edge approach that bridges the gap between theoretical DfS principles and their practical application, contributing to streamlined workflows, enhanced safety outcomes, and empowered decision making for construction professionals.