A keyword-priority genre classification scheme for metal-music corpus studies: eleven-bucket and seventeen-category encoding tables with a classified southern Chilean band corpus

This dataset documents a deterministic, keyword-priority classification scheme for the multi-label genre strings used by Encyclopaedia Metallum: The Metal Archives, the most comprehensive single-source repository of metal-music metadata currently available. The scheme provides researchers conducting corpus studies of metal music with a documented, replicable mapping from Encyclopaedia Metallum's free-text genre annotations to a fixed categorical vocabulary at two levels of granularity: An eleven-bucket primary scheme that assigns each band to one of eleven mutually exclusive categories (Grindcore/Gore, Black, Death, Doom/Sludge, Thrash/Speed, Heavy/Trad, Power, Folk/Pagan/Viking, Progressive, Metalcore/Deathcore, Industrial/Avant). A seventeen-category refinement that decomposes four of the eleven buckets (Black, Death, Doom/Sludge, Thrash/Speed) into finer sub-styles while preserving the remaining seven unchanged. Contents of the deposit: keywords_11bucket.csv — keyword priority table for the eleven-bucket scheme. keywords_17category.csv — keyword priority table for the seventeen-category refinement. classify.py — reference Python 3 implementation of the keyword-priority logic (no external dependencies). bands_classified.csv — classified development corpus of 560 metal bands from southern Chile (five administrative regions: La Araucanía, Los Ríos, Los Lagos, Aysén, Magallanes), 1988–2024. validation_subset.csv — inter-coder validation subset: 60 bands independently re-coded to the eleven-bucket level by a second domain expert, with automatic and manual assignments and an agreement flag. pilot_norway_classified.csv — out-of-sample portability demonstration: a 60-band random sample of Norwegian bands classified with the scheme unmodified. bands_input_template.csv and bands_classified_template.csv — example input and output formats. README.md — full documentation including methodology, rationale for priority ordering, usage instructions, and adaptation guidance for other regional metal-scene corpora. Reproducibility: Running classify.py on bands_input_template.csv reproduces bands_classified_template.csv exactly. The classification of the 560-band development corpus achieves a residual rate of 0.9% under the eleven-bucket scheme. Agreement between the automatic classification and an independent human coder on a 60-band stratified subset is 81.7% (Cohen's kappa = 0.78, 95% bootstrap CI [0.66, 0.88]). Reuse: The classification scheme is designed to be portable to other regional metal-scene corpora. An out-of-sample application to a 60-band Norwegian sample, with the keyword tables left unchanged, retains the scheme's core logic (residual rate 5.0%); researchers applying the scheme to corpora with substantially different stylistic profiles may need to augment the keyword tables for locally prominent styles. See the README for the detailed adoption workflow and the worked portability example. Ethics: The dataset consists of band-level metadata retrieved from the publicly available, fan-curated Encyclopaedia Metallum database. Band identifiers in the deposited files are pseudonymised through replacement with non-sequential numerical IDs; raw genre strings are preserved to enable replication and reuse. The development of the classification scheme was conducted within the framework of the ANID—FONDECYT Regular Project "Representaciones territoriales emergentes en la música popular del sur de Chile", which received ethical clearance from the Comité Ético Científico of Universidad de Los Lagos (Oficio Ordinario N° 025/2024, 23 April 2024). Related publication: A Data Descriptor manuscript documenting this dataset is currently under review at the journal Data (MDPI). The DOI of the published descriptor will be added to this record as a related identifier upon publication.