Going Beyond Keywords in Text

The exponential growth of digital information demands new approaches for identifying relevant documents efficiently. Keyword-based searches often fail to surface meaningful connections between text due to the locality of context. This paper emphasizes the role of semantic similarity analysis in improving relevance-based search. Moving beyond traditional keyword matching, we propose a system that leverages deep learning in text embedding models to capture both lexical and contextual nuances, thus, improving information retrieval. Using corpora of research publications, we explore how embedding models (e.g., Universal Sentence Encoder, fastText, Sentence Transformers) compare in finding relevant documents. Our results indicate that deep learning-based approaches, particularly Sentence Transformers, outperform traditional methods, demonstrating improved precision in handling long and complex queries. By focusing on semantic similarity analyses-based relevance, we show how this approach offers researchers a more context-aware tool for navigating large digital libraries.