Textual and structural approaches to detecting figure plagiarism in scientific publications

Abstract

The figures play important role in disseminating important ideas and findings which enable the readers to understand the details of the work. The part of figures in understanding the details of the documents increase more use of them, which have led to a serious problem of taking other peoples’ figures without giving credit to the source. Although significant efforts have been made in developing methods for estimating pairwise diagram figure similarity, there are little attentions found in the research community to detect any of the instances of figure plagiarism such as manipulating figures by changing the structure of the figure, inserting, deleting and substituting the components or when the text content is manipulated. To address this gap, this project compares theeffectiveness of the textual and structural representations of techniques to support the figure plagiarism detection. In addition to these two representations, the textual comparison method is designed to match the figure contents based on a word-gram representation using the Jaccard similarity measure, while the structural comparison method is designed to compare the text within the components as well as the relationship between the components of the figures using graph edit distance measure. These techniques are experimentally evaluated across the seven instances of figure plagiarism, in terms of their similarity values and the precision and recall metrics. The experimental results show that the structural representation of figures slightly outperformed the textual representation in detecting all the instances of the figure plagiarism.