<![CDATA[Measuring similarity between text documents is a fundamental task in Natural Language Processing (NLP) with broad applications, such as plagiarism detection, document clustering, and recommendation systems. This study aims to analyze and compare the performance of various document similarity algorithms, ranging from traditional lexical approaches to modern semantic methods. The algorithms reviewed include Bag-of-Words (BoW), TF-IDF, and Jaccard Similarity, as well as semantic representation-based methods such as Word Embeddings, Doc2Vec, and Sentence-BERT. An interactive web application was developed using the Gradio library to visualize comparison results in real-time and allow users to upload their own documents. The results indicate that lexical methods are effective at detecting keyword-based similarity but fail to capture semantic similarity when synonyms are used. Conversely, semantic methods—particularly Sentence-BERT—significantly outperform others in identifying contextual and semantic similarity, yielding more accurate scores for documents that differ in vocabulary yet share similar meanings. The study concludes that selecting the appropriate algorithm requires considering document characteristics and analysis objectives, and that the developed interactive tool can serve as an educational and experimental platform for such evaluations]]>
Copyrights © 2026
