Automated Schedule Enrichment from Daily Progress Reports: A Bi-Directional Neuro-Symbolic AI Approach

Abstract: Construction progress monitoring requires integrating daily progress reports (DPRs) with baseline schedules for productivity analysis and risk management. Manual integration is unscalable due to heterogeneous formats and inconsistent terminologies. This paper presents an automated system for enriching the schedules from the DPR files utilizing the large language models (neural) and knowledge graphs (symbolic) in a bidirectional integration, achieving the neuro-symbolic capability. It learns construction terminology from the schedule and the DPRs automatically, adapts its confidence weighting as domain knowledge matures, and validates against temporal, resource, spatial, and causal constraints. Tested on a real construction project with 173 DPR files and 1000 validation activities, results show a median confidence of 0.825, with zero constraint violations. It demonstrated substantial growth in the knowledge graph and greater robustness than pure neural and symbolic baselines. This neuro-symbolic integration offers a practical and interpretable solution for scalable schedule enrichment in real construction environments.