Retrieval-Augmented Generation for Culturally Nuanced Emotion Analysis in Nigerian Afrobeats Lyrics

This talk presents a retrieval-augmented generation system tailored for emotion analysis in Nigerian Afrobeats lyrics. Integrating supplementary dictionaries with a hierarchical classifier can enable the system retrieve culturally relevant context to refine emotion predictions. We discuss the architecture, implementation challenges and potential for addressing code-mixing and linguistic nuances in low-resource settings.

AI, LLMS, Music

In this talk, we introduce a novel retrieval-augmented generation (RAG) system designed to enhance emotion analysis in Nigerian Afrobeats lyrics. Our approach combines supplementary linguistic resources including Igbo-English, Yoruba-English, Pidgin-English, and English-Hausa dictionaries, with a hierarchical classifier. The system first uses a root-level classifier to assign a meta-emotion and then retrieves contextually relevant passages from a combined corpus of lyrics and dictionary entries. This retrieved context is appended to the original input, allowing the branch-level classifier to make more informed fine-grained predictions.

We will cover the system’s architecture, the process of building an enhanced retrieval corpus, and the integration of retrieval with hierarchical classification. The talk also discusses challenges we encountered such as data imbalance, error propagation from the root classifier, and ambiguity in dynamic emotion expressions, and outlines future work directions, including ensemble methods and explainability tools to further improve robustness and transparency. This work shows that using culturally informed resources can significantly improve emotion recognition in code-mixed, low-resource settings.