Enhancing Academic Integrity: An Analysis of Advanced Techniques for Plagiarism Detection using LESK, Word Sense Disambiguation, and SVM

Abstract

Plagiarism is widespread in academia, from ancient literature to modern research, where scholars' work is copied and published without authorization. In the late 90s, researchers explored various methods to detect plagiarism, including Word Sense Disambiguation (WSD), LESK, and Support Vector Machine (SVM). However, these conventional techniques have shown limitations in aligning with contemporary writing styles. This paper proposes an improved LESK algorithm for word sense detection and Improved SVM for feature extraction, addressing the shortcomings of existing methods and offering enhanced accuracy and efficiency in identifying plagiarized content. The study evaluates the proposed system using three datasets from PAN 2012, PAN 2013, and PAN 2014 documents to assess its performance across different types of text plagiarism. Results demonstrate the system's superiority, achieving higher classification accuracy when trained on the Second Dataset. A comprehensive analysis of the feature's significance in the training database reveals the importance of discriminative sentence similarity. The proposed system contributes to combating academic dishonesty, ensuring the authenticity of digital content in various contexts. Future work will explore cross-lingual plagiarism detection and image duplicity identification using Word Sense Disambiguation techniques. Additionally, efforts will be made to optimize time complexity for faster execution.