Towards Explainable Authorship Verification: An Approach to Minimise Academic Misconduct in Higher Education

Academic misconduct poses a growing challenge for higher education institutions worldwide. While AI presents valuable opportunities for learning enhancement, Unauthorized Content Generation (UCG) poses a significant threat to academic integrity. This paper addresses the challenges posed by UCG and explores innovative approaches to detection, focusing on the underutilised concept of authorship verification (AV). Despite the recognition of AV’s potential, its application in education has been limited. This study investigates the feasibility of utilising students’ academic writing profiles for AV to detect contract cheating and unacknowledged AI usage in academic contexts. Building upon previous research, this study enhances the existing Feature Vector Difference (FVD) AV method by introducing improvements to support better analysis, explainability, and interpretability of the classification process in an educational context. The refined classifier provides probability-based outputs, offering a transparent alternative to traditional “black box” binary outputs, and is able to identify stylometric features suitable for differentiating student’s writing profiles. Through this research, we contribute to the advancement of AV technology in education towards explainability, providing educators with a valuable tool to uphold academic integrity and combat the proliferation of UCG in educational environments.