Search academic texts

Information × Registration Number 0226U002927, (0125U001030) , R & D reports Title Software for information semantic search using large language models popup.stage_title Програмне забезпечення для семантичного пошуку інформації з використанням великих мовних моделей Head Zhuravchak Liubov M., Доктор технічних наук Registration Date 10-03-2026 Organization Lviv Polytechnic National University popup.description1 Improving methods and tools for creating software for information semantic search using an ontological approach and large language models. popup.description2 The work is devoted to the theoretical justification, development, and experimental research of specialized software for semantic information retrieval based on the use of large language models to form precise semantic representations of queries and documents. The research is based on the architecture of retrieval-augmented generation (RAG), which has been adapted to the specifics of Ukrainian-language academic databases and scientific publications in order to significantly improve the relevance of results. The system design process is based on a two-level vector search approach, which combines preliminary filtering by document annotations with subsequent detailed analysis of full-text segments. The proposed semantic search model uses Sentence Transformers for the Ukrainian language and specific vector similarity metrics, which allows for the effective extraction of the most significant text fragments from large arrays of unstructured data for optimal filling of the context of language models. A comparative analysis using the GPT-4-Turbo and Gemini 2.5 Flash models confirmed the superior effectiveness of the developed method, as it ensures the correct formation of the knowledge base, minimizes the risk of “hallucinations,” and increases the accuracy of responses to highly specialized queries. The practical significance of the results obtained lies in the possibility of their implementation in information retrieval systems, electronic libraries, and decision support systems, where high accuracy of knowledge extraction and minimization of information noise are critically important. Further research will focus on integrating semantic vector search with knowledge graphs and ontological modeling to provide a deeper level of logical inference and adaptive optimization of information ranking strategies. Product Description popup.authors Oleksandr K. Vyshnevskyy Denys I. Bratasiuk popup.nrat_date 2026-03-10 Close