OPTIMISATION OF AN ARTIFICIAL INTELLIGENCE MODEL FOR NEWS DATA CLASSIFICATION

DOI: 10.31673/2412-4338.2025.019647

Abstract

Abstract. This paper explores modern approaches to optimizing artificial intelligence models for news data classification, which is highly relevant in the fight against disinformation. The author analyzes existing methods of text processing and news classification, proposing a novel approach based on convolutional neural networks with class balancing techniques and L2 regularization. The main objective of the study is to improve news classification accuracy and reduce the impact of data imbalance on model training. Experimental studies conducted on the Fake and Real News dataset demonstrated the high accuracy of the optimized model, proving its effectiveness in distinguishing between fake and truthful news. The study analyzed the impact of various hyperparameters, including the number of convolutional layers, activation functions, regularization coefficients, and optimization algorithms. The results showed that applying class balancing techniques significantly reduces model errors and enhances its robustness against data distribution changes.The author also discusses prospects for further model improvement by integrating recurrent layers to account for sequential dependencies in text, which could significantly enhance its performance in real-world applications. Additionally, future research directions may include the application of transformer architectures, which have shown outstanding results in natural language processing tasks. Moreover, the possibility of using hybrid approaches that combine the advantages of convolutional and recurrent neural networks is considered, which could improve model efficiency. Implementing such methods will enable the development of more adaptive and universal solutions for text data analysis in the media domain.

Keywords: news classification, artificial intelligence, neural networks, model optimisation, disinformation