The growing volume of scientific literature makes it increasingly difficult to retrieve relevant information efficiently, particularly in specialized domains such as lunar dust research. Traditional information retrieval systems are not equipped to manage the scale and complexity of technical content found in niche scientific fields. This study presents a machine learning approach that leverages pre-trained Large Language Models (LLMs), fine-tuned on domain-specific data, to improve the accuracy and contextual relevance of automated tagging. The model was evaluated using a dataset of 189 lunar dust-related conference papers from NASA’s NTRS repository, each containing abstracts, full text, and author-provided keywords. Tags and summaries generated by the model were assessed against these references using semantic similarity scoring and frequency analysis. Following the CRISP-DM methodology, the study compares this approach to rule-based systems, Bag-of-Words, Latent Dirichlet Allocation, and traditional classifiers. These alternatives were found to lack the scalability and contextual understanding needed for accurate tagging in complex scientific literature. The study also analyzes how dataset size and domain specificity influence model performance, with results demonstrating clear benefits from targeted fine-tuning. This work offers a scalable solution for automating the categorization of technical literature and supports more effective information retrieval in fields where precision and domain expertise are critical.