Improving the accuracy of schedule information communication between humans and data

Construction schedules are written instructions of construction execution shared between stakeholders for essential project information exchange. However, construction schedules are semi-structured data that lack semantic details and coherence within and across projects. This study proposes an ontology-based Recurrent Neural Network approach to bi-directionally translate between human written language and machinery ontological language. The proposed approach is assessed in three areas: text generation accuracy, machine readability, and human understandability. This study collected 30 project schedules with 19,589 activities (sample size = 19,589) from a Tier-1 contractor in the UK. The experimental results indicate that: (1) precision and recall of text generation LSTM-RNN model is 0.991 and 0.874, respectively; (2) schedule readability improved by increasing the semantic distinctiveness, measured using the cosine similarity which was reduced from 0.995 to 0.990 (p < 0.01); (3) schedule understandability improved from 75.90% to 85.55%. The proposed approach formalises text descriptions in construction schedules and other construction documents with less labour investment. It supports contractors to establish knowledge management systems to learn from historic data and make more informed decisions in future similar scenarios.

Automated summary

This study proposes an ontology-based Recurrent Neural Network approach to bi-directionally translate between human written language and machinery ontological language. Experimental results show that the proposed approach has good performance in text generation accuracy, machine readability, and human understandability.