ORIN Lyrics Dataset: A Comprehensive Corpus of Multilingua Nigerian Song Lyrics for NLP
Description
The ORIN Lyrics dataset provides a comprehensive corpus of song lyrics from Nigeria, spanning diverse languages (Yoruba, English, Nigerian Pidgin, and multilingual combinations), genres, and styles prevalent in Nigerian popular and traditional music. The dataset contains 862 carefully curated entries, each accompanied by detailed metadata including Artist, Title, Album, Year, Genre, Subgenre, Region, Language, Group, and Source URLs. Compiled through a hybrid method involving web scraping from popular music platforms and subsequent manual verification and metadata enrichment, ORIN Lyrics is tailored specifically for research purposes, particularly in natural language processing (NLP), sentiment analysis, music information retrieval (MIR), computational ethnomusicology, and cultural AI applications. This dataset is particularly valuable for researchers and practitioners working with low-resource languages, cultural text mining, and cross-lingual NLP tasks. This dataset is used in the study Green Topics, Deep Roots: Energy-Aware Topic Modelling of Multilingual Nigerian Lyrics. The files include raw lyric text, basic metadata (ID, artist, title, language tags), preprocessed text used for modeling, and model/evaluation results (coherence C_v and U_mass, topic diversity, inter-topic Jaccard overlap, and logged energy consumption in kWh for model runs). The dataset supports reproducibility of energy-aware benchmarking of topic modelling methods (NMF, LDA, LSI, HDP, BERTopic, Top2Vec, GSDMM) and includes notes on preprocessing and limitations. See README and DATA_DESCRIPTOR for full details (collection, preprocessing, evaluation metrics, and limitations). Key contents: raw lyrics, preprocessed text, metadata, topic modelling results, and energy logs. Limitations: Drawn from publicly available Nigerian lyrics; analysis focuses on English–Yoruba–Pidgin lines and uses intrinsic topic metrics and a single hardware profile for energy logging. See the paper and DATA_DESCRIPTOR for details and caveats. If you use this dataset, cite it like this: Folorunso, S. O., Sina, A. T., Oladipo, F. O., & Rukayat, O. G. (2025). Green Topics, Deep Roots: Energy-Aware Topic Modelling of Multilingual Nigerian Lyrics. In The Thirty-ninth Annual Conference on Neural Information Processing Systems (NeurIPS), Creative AI Track: Humanity.
Files
Steps to reproduce
The ORIN Lyrics dataset was assembled using a combination of web scraping techniques (implemented via Python with BeautifulSoup and Requests libraries) and meticulous manual verification. The initial raw data collection involved scraping song lyrics and associated metadata from popular music platforms like Boomplay and Spotify. The manual verification process ensured accuracy and completeness, including metadata enrichment and quality assurance by domain experts familiar with Nigerian music and languages. Data Format: The dataset is available as a structured Excel (.xlsx) file with clearly labeled columns: S/N: Record identifier. Artist: Name of the performer or band. Lyrics: Full text of song lyrics. Title: Song title. Album: Album name (where available). Year: Year of release. Genre: Primary music genre classification. Subgenre: Additional detailed genre classification. Source: URLs linking to the original lyrics source. Time: Duration of the song in "HH:MM:SS" format. Region: Region/country of origin. Language: Language or multilingual combinations. Group: Broader musical category or grouping.
Institutions
- Olabisi Onabanjo University