TECHNOLOGY FOR ANALYZING TEXT TONES USING A CONVOLUTIONAL NEURAL NETWORK

Remarchuk Valery Nikolaevich, Goryachkin Boris Sergeevich, Gvozdeva Yana Vyacheslavovna, Malina Sofya Vladimirovna

Problem statement. Despite the existence of numerous sentiment analysis methods, their accuracy and effectiveness remain insufficient for solving complex tasks such as analyzing political and economic texts, predicting market trends, or ensuring security. Traditional approaches, including dictionary-based methods (e.g., SentiWordNet, AFINN) and combined techniques, demonstrate limited accuracy (72.7-78.2%), making them unsuitable for tasks requiring high reliability. Moreover, these methods perform poorly in processing contextual dependencies, irony, sarcasm, and other complex linguistic constructs frequently encountered in texts. Goal. Automatisation of sentiment analysis by developing corresponding software solutions to improve the objectivity of text analysis. Results. CNN-based technology of sentiment analysis has been developed. 81.64% accuracy has been achieved, superior to traditional techniques. Practical significance. Practical significance of this study lies in its potential to enhance the quality of existing automated opinion extraction systems through the implementation of convolutional neural network (CNN)-based technology. These advancements can be directly applied to sentiment analysis systems that aggregate and process data in real-time across various platforms including social networks, customer reviews, and news sources. Proposed methodology enables more effective monitoring of sentiment patterns, which proves particularly valuable for detecting intentionally misleading content. By analyzing deviations in sentiment between entity pairs and comparing them against established baseline metrics within specific timeframes, the system can identify textual claims that demonstrate statistically significant anomalies, potentially indicating representations that diverge from factual reality. This capability holds particular relevance for applications requiring high-fidelity sentiment analysis, such as media monitoring, reputation management, and information verification systems, where distinguishing authentic sentiment from manipulated content becomes increasingly crucial in the digital information ecosystem. The CNN-based approach demonstrates superior performance in capturing nuanced linguistic patterns and contextual relationships compared to traditional sentiment analysis methods, thereby offering more reliable results for critical decision-making processes.

Keywords:Sentiment analysis, text, automatisation of sentiment analysis, machine learning, Convolutional Neural Network