Leveraging NLP and Expert Knowledge Integration for Enhanced Project Prediction and Dynamic Task Management in Multidisciplinary Engineering

Academic Institution: University of Strathclyde

Academic Supervisor: Dr Andy Wong

PhD Student: Daniel Williamson

Industrial Partner: Morrison Construction Ltd

Summary

Construction projects involve many complex and interconnected tasks that can lead to delays or inefficiencies, especially when considering sustainability goals. The aim of this project is to develop an intelligent scheduling system that utilises advanced machine learning techniques, including Natural Language Processing (NLP) and Graph Neural Networks (GNN), to predict and optimise the completion timelines of these tasks.

The system will continuously learn from project reports and update its intelligence to improve its ability in prediction and optimisation, helping project managers make more informed decisions that enhance both project efficiency and sustainability.

The aim of the project is to develop a system that can utilise expert knowledge to address inefficiencies in task scheduling and resource allocation by understanding how tasks are interdependent and how these dependencies evolve throughout the project lifecycle.

To tackle this, the system integrates a hybrid model composed of a Convolutional Neural Network (CNN) and Long Short-Term Memory (LSTM) units to predict project timelines. Additionally, Natural Language Processing (NLP) is employed to analyse monthly project reports, identifying key changes in task dependencies and other contextual information. A Graph Neural Network (GNN) is then used to model these evolving relationships between tasks as a dynamic graph, where nodes represent tasks and edges represent dependencies that change over time.

The NLP-GNN component is designed to extract and model the evolving relationships between tasks as described in project reports. As task dependencies can change from month to month—due to delays, new tasks, or resource allocation adjustments—the system needs a mechanism to continuously learn and adapt to these changes. This component provides a dynamic graph representation of the project’s task structure, where tasks are represented as nodes and dependencies between tasks as edges.

The CNN-LSTM component is used for time series prediction of project timelines, helping to forecast when tasks are expected to be completed. While the NLP-GNN component models task dependencies, the CNN-LSTM manages the temporal sequence of tasks, predicting future outcomes based on past task performance and dependency changes.

Predictions from the CNN-LSTM model feed back into the NLP-GNN component, adjusting the task dependency graph based on predicted delays or changes in task priorities. This operates as a continuous learning loop, where the system becomes more accurate as more data is processed, ultimately leading to optimised project timelines and improved task scheduling.

The graph structure will provide more accurate predictions of task completion times, adjusting for changes in task interactions. It will adapt continuously from historical project data and updates, allowing for varying project conditions. By incorporating both machine learning-based predictions and expert knowledge constraints, the system will support project managers in making data-driven decisions that improve project timelines and sustainable outcomes.

Key Results/Outcomes

  • Design of a database linking key attributes and dependencies for task management to textual statements.

  • The design of a BagofWords model refined on Expert Knowledge terms and Corresponding Outputs.

  • The ability to fine-tune a dynamic GNN using Textual Attributes and Time Series Outputs.

  • Demonstrated through a company case study, the ability of the system to produce Dynamic adjustments to task dependencies based on real-time project changes or delays, providing up-to-date recommendations.

  • The examination of the system and user trust to improve stakeholder acceptance.

  • Comparison between the predicted vs. actual user performance to highlight areas of strength and need for improvement.

Contact Details

Dr Tse Chiu (Andy) Wong

Senior Lecturer (Associate Professor) in Engineering Management | Head of Engineering Management Research Group, University of Strathclyde

Email: andy.wong@strath.ac.uk

Daniel Williamson

PhD Student, University of Strathclyde

Email: daniel.williamson@strath.ac.uk