Computational Approaches to Sentiment Analysis in Nigerian Social Media Text: A Corpus-Linguistic and NLP Perspective

Sentiment analysis of social media text has become a major application domain for natural language processing, yet existing computational sentiment tools perform significantly below their reported accuracy levels when applied to Nigerian social media text due to the prevalence of code-mixing, Nigerian Pidgin, indigenized English idioms, and culturally specific sentiment expression conventions not represented in training data. This dissertation develops computational approaches to sentiment analysis specifically adapted for Nigerian social media text, making original contributions to both computational linguistics and corpus-based sociolinguistics. The dissertation's contribution operates at three levels: a corpus construction contribution involving the development of a two-million-word annotated Nigerian social media corpus sampling Twitter, Facebook, and WhatsApp broadcast text with sentiment, language variety, and code-mix labels; a linguistic analysis contribution providing the first systematic computational description of sentiment expression conventions in Nigerian Pidgin and code-mixed Nigerian English text; and a model development contribution constructing and evaluating three computational sentiment analysis architectures adapted for Nigerian multilingual input. Deep learning models incorporating transfer learning from multilingual pre-trained language models, augmented with task-specific Nigerian linguistic feature engineering, are compared against standard baseline sentiment classifiers across test sets sampled from the corpus. Results demonstrate that architecture incorporating Nigerian Pidgin word embeddings and code-mix boundary features achieves significantly higher accuracy on Nigerian social media test sets than state-of-the-art English-only models. The dissertation makes methodological contributions applicable to other multilingual African social media language processing contexts. Keywords: sentiment analysis, Nigerian social media, computational linguistics, code-mixing, NLP.