Analyzing Similarity in Mathematical Content To Enhance the Detection of Academic Plagiarism

TL;DR

This paper reviews approaches for analyzing mathematical content to improve academic plagiarism detection, highlighting current methods' strengths and limitations, and proposing future research directions for more comprehensive detection.

Contribution

It provides an overview of mathematical information retrieval techniques and analyzes their potential for detecting various forms of mathematical plagiarism, including disguised cases.

Findings

Syntax-based approaches excel at detecting undisguised plagiarism.

Structure-based and hybrid approaches can identify disguised plagiarism with renamed identifiers.

Current methods are limited to formula-level detection; section-level and equivalence transformations require further research.

Abstract

Despite the effort put into the detection of academic plagiarism, it continues to be a ubiquitous problem spanning all disciplines. Various tools have been developed to assist human inspectors by automatically identifying suspicious documents. However, to our knowledge currently none of these tools use mathematical content for their analysis. This is problematic, because mathematical content potentially represents a significant amount of the scientific contribution in academic documents. Hence, ignoring mathematical content limits the detection of plagiarism considerably, especially in disciplines with frequent use of mathematics. This paper aims to help close this gap by providing an overview of existing approaches in mathematical information retrieval and an analysis of their applicability for different possible cases of mathematical plagiarism. I find that whereas syntax-based approaches perform particularly well in detecting undisguised plagiarism, structure-based and hybrid approaches promise to also detect forms of disguised mathematical plagiarism, such as plagiarism with renamed identifiers. However, more research in this area is needed to enable the detection of more complex mathematical plagiarism: the scope of current approaches is restricted to the formula-level, an extension to the section-level is needed. Additionally, the general detection of equivalence transformations is currently not feasible. Despite these remaining problems, I conclude that the presented approaches could already be used for a basic automated detection system targeting mathematical plagiarism and therefore enhance current plagiarism detection systems.