Towards trustworthy AI for link prediction in supply chain knowledge graph: a neurosymbolic reasoning approach

Edward Elson Kosasih,Alexandra Brintrup

DOI: https://doi.org/10.1080/00207543.2024.2399713

IF: 9.018

2024-10-05

International Journal of Production Research

Abstract:Modern supply chains are complex and interlinked, resulting in increased network risk exposure for companies. Digital Supply Chain Surveillance (DSCS) has emerged as a practice to proactively monitor these risks. Within DSCS, link prediction, a particular task whose objective is to identify hidden relationships in the supply chain, has been increasingly studied in the literature. While many approaches have been proposed, machine learning (ML) based techniques have recently gained wider attention due to their state-of-the-art performance across different benchmark datasets. However, adoption of these technologies in practice remains difficult. Their black-box nature have resulted in a lack of trustworthiness among practitioners. In this paper, we design a trustworthy ML approach based on recent theoretical development in neurosymbolic AI methods that enables a more transparent learning and reasoning process. We find that our approach is not only on par with state-of-the-art black box models when tested in two benchmark datasets on the automotive and energy industry but also addresses research gaps on explainability and complexity analysis.

engineering, manufacturing, industrial,operations research & management science