evelopment of a methodology for increasing the efficiency of processing unstructured text information in defense information, information and communication systems

The article provides a comprehensive analysis of modern methods for processing unstructured text information, including: linguistic, statistical, vector models of text representation, machine learning and deep learning, sequential analysis, and identifies their advantages and limitations in the conditions of functioning of information and analytical systems for defense purposes. It is shown that in the conditions of increasing volumes of information flows and the need for their processing in real time, most traditional approaches are characterized by significant requirements for computing resources and delays in decision-making, which is due to the need for a complete analysis of unstructured text information. It is established that existing methods do not provide a proper balance between classification accuracy and processing speed at the stage of primary processing and ensuring the ability to further in-depth analysis at subsequent stages, which is critical for decision-making support systems in the military sphere. A method based on a hybrid approach is proposed that provides increased efficiency in text data processing through the use of multi-alternative sequential decision rules at the initial processing stage and transformer models for subsequent in-depth analysis, which provides the possibility of early decision-making at the preliminary stage and on an incomplete sample. This approach allows you to reduce the amount of processed data and reduce the analysis time without significantly reducing the reliability of the results. Unlike classical classification methods, the proposed solution provides flexible management of the process of processing unstructured input text information. The use of Data Lake type storages is proposed to integrate heterogeneous information sources and ensure system scalability. It is substantiated that the implementation of the proposed method allows you to increase the efficiency of text information processing, reduce computational costs and ensure the required level of classification accuracy in conditions of uncertainty and information overload. The results obtained can be used in the creation and modernization of information and analytical systems for defense purposes, as well as in integration with existing data processing platforms within the framework of modern military standards.