A Latent Semantic Analysis (LSA) approach for plagiarism detection in Arabic documents

Abstract

Plagiarism detection in Arabic text is complicated by the language’s morphological richness, where surface-level word matching often fails to capture semantic similarity between paraphrased passages. This work applies Latent Semantic Analysis to represent documents in a reduced semantic space, allowing detection of conceptually similar content even when the wording differs.

Method

Documents are represented as term-document matrices, then projected into a lower-dimensional semantic space using singular value decomposition, so that similarity can be measured on meaning rather than exact word overlap.

Contribution

Shows that LSA based semantic similarity outperforms purely lexical matching approaches for detecting plagiarism in Arabic documents, particularly for paraphrased or reworded content.