Analytical Examination of Natural Language Processing Performance for Hausa Language Text Classification in Nigerian Social Media Contexts

Natural language processing tools and models trained predominantly on English and European languages perform poorly on low-resource African languages such as Hausa, and evaluating NLP performance for Hausa text classification in Nigerian social media contexts fills a significant technical and linguistic research gap. This study analytically examines the performance of NLP approaches for Hausa language text classification in Nigerian social media environments. A computational experimental design was applied using a curated dataset of 25,000 Hausa-language Twitter and Facebook posts from Nigerian users, manually labelled across five categories: news, political opinion, commercial advertisement, health information, and misinformation. Six text classification approaches were evaluated: Naive Bayes with TF-IDF features, support vector machine, bidirectional LSTM, fine-tuned multilingual BERT, Hausa-adapted AfroXLM-R, and a hybrid ensemble model. Classification performance was measured using macro-averaged F1-score, precision, and recall. Available Hausa NLP literature from Nigerian computational linguistics identifies code-switching between Hausa and English, dialectal variation between Kano and Sokoto Hausa, and absence of large labelled Hausa NLP datasets as the primary technical challenges. The Transfer Learning Framework and the Low-Resource NLP Adaptation Theory provide the analytical basis. Findings indicate that AfroXLM-R fine-tuned on the study dataset achieves the best macro-F1 score of 0.824, substantially outperforming traditional ML approaches. The study fills a performance benchmarking gap in Nigerian Hausa NLP research. Recommendations address open Hausa NLP corpus development and localised model sharing through NLP Nigeria community. Keywords: natural language processing, Hausa language, text classification, social media, Nigeria.