Hierarchical Article Classification: A Multi-Level Framework for Organizing Scholarly Literature

The classification of scholarly publications is a critical task for knowledge organization, yet traditional methods often fall short in addressing the complexities of interdisciplinary research and evolving hierarchical structures. This study introduces a novel Hierarchical Article Classification (HAC) framework that leverages machine learning and deep learning techniques to improve accuracy, adaptability, and scalability in classifying scientific documents across multiple hierarchical levels. The proposed framework integrates text transformation methods, hierarchical classification strategies, classical and cutting-edge machine learning models, including SVM and transformer-based architectures like BERT and GPT-4, and evaluation metrics. A comprehensive experimental analysis conducted on a dataset from Fiocruz demonstrates that the framework is particularly effective and is able to identify models that are especially accurate in the Biological and Health Sciences fields. The main contributions of this work include: (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i</i>) a modular framework capable of handling incomplete data and hierarchical structures, adaptable to diverse domains, (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ii</i>) a practical application using Fiocruz’s scientific publications, and (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">iii</i>) a robust evaluation methodology combining traditional metrics, hierarchical metrics, and expert analysis. The results illustrate the potential of integrating deep learning and contextual embeddings for sensitive, comprehensive, and automated literature organization.