Hybrid Plagiarism Detection using Ensemble Similarity and Intrinsic Stylometric Analysis

Plagiarism in student submissions poses a serious threat to academic integrity. This study focuses on detecting such plagiarism using Natural Language Processing (NLP) and Machine Learning (ML) techniques. Unlike traditional tools that detect only exact matches, this approach aims to identify paraphrased and semantically similar content. The proposed system utilizes a hybrid approach: a weighted ensemble of four extrinsic similarity metrics (Semantic, TF-IDF, Jaccard, and LCS) combined with a Novel Intrinsic Stylometric Consistency Score (ISCS). The ISCS is designed to detect internal stylistic inconsistencies, which are indicative of content splicing or patch-writing (intrinsic plagiarism), even without a source document. The proposed system improves the detection accuracy of obfuscated plagiarism and supports educational institutions in maintaining originality.