ADVANCING ENTERPRISE-SCALE INFORMATION RETRIEVAL AND COMPREHENSION USING LARGE LANGUAGE MODELS

The article analyses the implementation of natural language query processing systems for automating information retrieval in enterprises from the perspective of modern information technologies and data science.The study's relevance is determined by the need to optimize access to large volumes of data and enhance the efficiency of management processes by integrating intelligent systems for processing user queries.It has been established that such systems facilitate quick access to relevant information by interpreting natural language queries and generating structured responses.However, problems related to the contextual understanding of queries, multilingual processing, adaptation to domain-specific terminology, and maintaining performance under heavy loads have been identified.The article aims to analyze the key issues in the operation of natural language query processing systems and develop recommendations for improving their efficiency to optimize information processes in enterprises.Methodology.The research is based on a systematic analysis of query processing algorithms, an evaluation of language model performance, and a comparative analysis of accuracy, completeness, and query execution time for queries of varying complexity.Risk analysis methods were used to identify key issues and build optimization models.Scientific novelty.The article comprehensively analyzes the applied aspects of using language models for natural language query processing.The main system limitations have been identified, and measures have been proposed to improve response accuracy and processing speed.Special attention is given to algorithms for automatic query clarification and recommendations for adapting language models to domain-specific data.Conclusions.It has been proven that implementing natural language query processing systems improves the efficiency of information retrieval, provided algorithm transparency and personalization mechanisms are ensured.Recommendations have been made for training models on domain-specific texts, implementing data aggregation algorithms, and using hybrid solutions to optimize query processing in multi-user environments.