Advanced Engineering Informatics, vol.62, 2024 (SCI-Expanded)
The escalating volume of data in engineering practice necessitates innovative computational approaches for data-driven insights. Existing literature relies on isolated data points, unable to exploit the inherent connectivity in engineering datasets, resulting in suboptimal utilization of data context. This research employs node2vec, a graph-based recommendation system that surpasses existing models by incorporating an efficient walking mechanism to learn from past behaviors and a predictive component that enhances its adaptability. By structuring these activities into a network of budgeted units, person-hours, and earned values, the effectiveness of the node2vec model as a resource recommendation tool was demonstrated across three diverse datasets. Firstly, node2vec efficiently explores diverse neighborhoods within the input network through a flexible biased random walk, enhancing the system's ability to adaptively model complex relationships among various project elements. Secondly, this graph-based approach allows the recommendation models to fully harness relational data. These mechanisms coupled with a predictive neural network component enabled node2vec to learn from and utilize data connectivity, achieving an accuracy rate of 94% across the datasets. Ultimately, by leveraging collected engineering data and recognizing dependencies among records, the system can offer more detailed insights and empower engineering managers to make better-informed decisions.