SEMANTIC TEXT SIMILARITY DETECTION USING TRANSFORMER-BASED DEEP LEARNING MODELS FOR ADVANCED PLAGIARISM IDENTIFICATION

Abstract

The rapid growth of digital information resources and online academic content has significantly increased the need for effective plagiarism detection systems. Traditional plagiarism detection approaches primarily rely on lexical and syntactic similarities, making them less effective in identifying paraphrased or semantically modified texts. This study investigates the application of Transformer-based deep learning models for semantic text similarity detection to enhance advanced plagiarism identification. The proposed approach employs modern Natural Language Processing (NLP) techniques, including text preprocessing, semantic embedding generation, and similarity measurement using cosine similarity. Transformer-based models such as BERT and Sentence-BERT are utilized to capture contextual and semantic relationships between texts beyond surface-level word matching. Experimental evaluation demonstrates that semantic embedding models outperform conventional methods by accurately detecting hidden semantic similarities in paraphrased documents.