Enhancing amharic news classification through ontology-based feature fusion and logistic regression.

Amharic, a major Semitic language of Ethiopia, remains underrepresented in natural language processing research due to limited linguistic resources. This study addresses the challenge of accurate text classification for low-resource languages by proposing a hybrid framework that integrates a semantically structured Amharic News Ontology (ANO) with traditional TF-IDF features. The ANO was systematically developed through a rigorous four-phase methodology to capture hierarchical relationships between news domain concepts. We formalize a feature fusion technique that combines lexical (TF-IDF) and ontological features into enriched document representations used to train a Logistic Regression classifier. Evaluated on a public dataset of 61,915 Amharic news articles across six categories, our ontology-enhanced model achieved 97.0% accuracy. Statistical analysis revealed a 3.2 percentage point improvement over a TF-IDF-only baseline (McNemar's test ([Formula: see text], [Formula: see text])), with ontology integration particularly effective in disambiguating semantically related categories (Politics-Business confusion reduced by 38%). The experimental results suggest that, for Amharic news classification on the evaluated dataset, integrating structured semantic knowledge through a domain-specific ontology improves classification accuracy by 3.2% compared to a TF-IDF baseline. While these findings demonstrate the potential of ontology-based feature fusion for low-resource languages, they are constrained to the news domain and the specific dataset used.