Node Embeddings and Graph Representation Learning to Capture Structural and Semantic Information on Causal Relationships within Pre-Construction Delay Networks

This study proposes a framework based on graph representation learning to model pre-construction risks as part of an interconnected network and determines the root causes behind these delays. A list of 30 construction delays was identified and validated by a panel of experts. The dependencies between the different risk factors were quantified using the analytic hierarchy process (AHP), allowing for the construction of a risk network as well as risk owner and risk domain interface matrices. Combining topological metrics and the developed risk network, node embeddings were computed using the Node2Vec algorithm to identify the critical elements from each of the risk owner and risk domain interface. Results show that the project team and contractual risks are the most critical elements from the risk owner and risk domain interfaces, respectively. This study provides guidance to project managers for developing more robust risk management strategies and allocating their resources more effectively.