Text Tone Analysis in Social Networks: Research, Evaluation and Comparison of Algorithm Performance

The article is devoted to a comprehensive study of methods for analyzing the tonality of textual information in order to improve the effectiveness of information security monitoring and public opinion management in the digital environment. The authors conduct a comparative analysis of approaches based on lexicographic methods, machine learning with a teacher (naive Bayesian classifier and maximum entropy method) and systems using a set of rules. The study includes a detailed review of the algorithmic features, advantages and limitations of each method, as well as an evaluation of their adaptability when the topics of the analyzed texts change. The results of the comparative analysis show that machine learning methods demonstrate higher performance in terms of accuracy, completeness and F-measure compared to rule set based systems. The authors also draw attention to the dependence of the results on the specifics of the training sample and the peculiarities of the contextual application of algorithms. The practical significance of the work lies in the possibility of using the developed methods for the rapid detection of negative content and predicting changes in classification accuracy when adapting to new topics, which is especially relevant for information security and management decision support in a dynamically developing digital environment.