Implementation of neural networks for the classification of information sources in digital journalism about artificial intelligence

Torres-Cruz F.; Yucra-Mamani Y.J.; Aragón-Cruz W.E.; Tovar-Yucra M.E. · European Public and Social Innovation Review (2025) · DOI: 10.31637/epsir-2025-1436

Resumen

Introduction: In the digital era, classifying information sources is crucial for maintaining the quality of journalism, especially with artificial intelligence (AI). This study employs neural networks for this task, evaluating their effectiveness and providing clear interpretations of the results. Methodology: A dataset with 14 characteristics of journalistic content was used, including genre, publication section, source type, and multimedia presence. The target variable classified the primary source of the text into categories such as expert, political, cultural, religious, journalistic, and others. The neural network model had two hidden dense layers with 64 neurons each, using ReLU activation. It was trained and evaluated with data split into training and testing sets. Feature standardization improved the model’s performance, achieving 46% accuracy in testing. Results: Techniques like SHAP and LIME were applied to interpret the model’s predictions. SHAP identified the most influential features. LIME provided a detailed understanding of how specific features affect predictions. Conclusions: This study proposes an innovative approach to classifying information sources in digital journalism and highlights the importance of interpretability in AI models.

Instituciones: Universidad Nacional del Altiplano; Universidad Católica de Santa María