Visualizing document similarity using n-grams and latent semantic analysis

As the number of information resources and document quantity explodes, efficient tools with intuitive visualization capabilities desperately needed to assist users in conducting document similarity analysis and/or plagiarism detection tasks by discovering hidden relations among documents. This paper proposes a content-based method for document similarity analysis and visualization. The proposed method is based on modeling the relationship between documents and their n-gram phrases, which are generated from the normalized text, exploiting morphology analysis and lexical lookup. Resolving possible morphological ambiguities is carried out by tagging the words within the examined documents. Text indexing and stop-words removal are performed, employing a new technique that is efficient in dealing with multiple long documents. The examined documents' TF-IDF model is constructed using heuristic based pair-wise matching algorithm, considering lexical and syntactic changes. Then, the hidden associations between the documents and their unique n-gram phrases are investigated using Latent Semantic Analysis (LSA). Next, the pairwise document subset and similarity measures are derived from the Singular Value Decomposition (SVD) computations. Different visualization techniques are then applied on the SVD results to expose the hidden relations among the documents under consideration. As Arabic is one of the most morphological and complicated languages, this paper emphasizes Arabic documents similarity analysis and visualization. Various experiments were carried out revealing the strong capabilities of the proposed method in analyzing and visualizing literal and some types of intelligent similarities.