CoolFace
Datasetpublic

baiango/genre-taxonomy-nsfw

Genre Taxonomy — NSFW Split Content advisory: this is a taxonomy of genre/subgenre labels (no prose), but the labels denote adult, sexually explicit, and graphic-violence themes. Intended for content-filtering, routing, and dataset-hygiene research. The adult half of a two-part short-fiction genre taxonomy: 2,194 genre entries / 3,940 labels (genre + subgenre names; no story text). The safe-for-work companion split lives in the paired repo genre-taxonomy-sfw. Every label was… See the full description on the dataset page: https://huggingface.co/datasets/baiango/genre-taxonomy-nsfw.

sourceHugging Facecc0-1.0updated 13d agoView on Hugging Face
2likes52downloads
Dataset Card

Genre Taxonomy — NSFW Split

Content advisory: this is a taxonomy of genre/subgenre labels (no prose), but the labels denote adult, sexually explicit, and graphic-violence themes. Intended for content-filtering, routing, and dataset-hygiene research.

The adult half of a two-part short-fiction genre taxonomy: 2,194 genre entries / 3,940 labels (genre + subgenre names; no story text). The safe-for-work companion split lives in the paired repo genre-taxonomy-sfw.

Every label was screened by multi-round LLM judging (large single-pass scan, then targeted re-judging rounds) under a double-pass agreement standard: a label is only acted on when independent judging passes agree. Labels judged sexual/adult or graphic-violence were routed here from the SFW split; labels judged to sexualize minors were removed from the taxonomy entirely (banned in both splits — they do not appear in this file either).

Schema

columntypedescription
genrestringparent genre label
subgenreslist\<string\>child labels under the parent

Provenance

  • —Expanded from an LLM-generated taxonomy (multi-teacher distillation pipeline).
  • —Content adjudication frozen 2026-09-12 (v32); multi-round judge audit closed with zero unresolved flags.
  • —Label-hygiene pass 2026-09-12 (v33): removed 1 corrupted list-named genre entry and stripped stray punctuation from 44 labels. No label content was re-judged or reclassified.
  • —Source snapshot md5: c8ccb5cdd5883a6ea13c34256a9f358a (genres_nsfw.json, v33); this repo ships the same data as tidy parquet with internal bookkeeping stripped.

Notes

  • —Labels only — this dataset contains no prose, instructions, or story content.
  • —CC0: released to the public domain.