---
task_categories:
- text-classification
language:
- en
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: dev
    path: data/dev-*
  - split: test
    path: data/test-*
  - split: validation
    path: data/validation-*
dataset_info:
  features:
  - name: text
    dtype: string
  - name: label
    dtype: string
  - name: label_id
    dtype: int64
  - name: gender
    dtype: string
  - name: group_id
    dtype: int64
  - name: author_id
    dtype: string
  - name: source_id
    dtype: string
  - name: job
    dtype: string
  - name: age
    dtype: int64
  - name: age_band
    dtype: string
  - name: sign
    dtype: string
  splits:
  - name: train
    num_bytes: 42000916
    num_examples: 25956
  - name: dev
    num_bytes: 5995268
    num_examples: 3755
  - name: test
    num_bytes: 11989095
    num_examples: 7425
  - name: validation
    num_bytes: 5995268
    num_examples: 3755
  download_size: 39778969
  dataset_size: 65980547
---

# BlogTopic

Prepared for gender-fairness (TPR-gap) evaluation of fine-tuned language models.

## Schema

| column | type | meaning |
|---|---|---|
| `text` | str | input text |
| `label` | str | task class name (Y), used as the generation target in prompted fine-tuning |
| `label_id` | int | contiguous class id (Y) |
| `gender` | str | protected attribute (Z), `female` or `male` |
| `group_id` | int | 0 = female, 1 = male |

Splits: `train`, `dev` (aliased as `validation`), `test`.

## Key statistics

- Examples: 37,136
- Classes: 26
- Test P(female|Y) range: 0.186 - 0.792
- Test Cramer's V (Y vs Z): 0.309
- Smallest test Y x Z cell: 21

Full per-class Y x Z tables are in `dataset_stats.json`.

## Provenance

- **raw_path**: /projects/u6cg/sshao/code/peft/rep_under_control/data/raw/blog/data/blogs.zip
- **files_seen**: 19320
- **files_skipped_bad_filename**: 0
- **bloggers_skipped_unknown_industry**: 6827
- **unit**: post
- **raw_industry_counts**: {'indunk': 6827, 'student': 5120, 'education': 980, 'technology': 943, 'arts': 721, 'communications-media': 479, 'internet': 397, 'non-profit': 372, 'engineering': 312, 'government': 236, 'law': 197, 'consulting': 191, 'science': 184, 'marketing': 180, 'businessservices': 163, 'publishing': 150, 'advertising': 145, 'religion': 139, 'telecommunications': 119, 'military': 116, 'banking': 112, 'accounting': 105, 'fashion': 98, 'humanresources': 94, 'tourism': 94, 'transportation': 91, 'sports-recreation': 90, 'manufacturing': 87, 'architecture': 69, 'chemicals': 62, 'biotech': 57, 'lawenforcement-security': 57, 'realestate': 55, 'museums-libraries': 55, 'construction': 55, 'automotive': 54, 'agriculture': 36, 'investmentbanking': 33, 'environment': 28, 'maritime': 17}
- **cleaning**: {'input_rows': 76872, 'dropped_unusable_gender': 0, 'dropped_too_short': 6736, 'dropped_duplicate_text': 1713, 'output_rows': 68423}
- **label_field**: industry
- **carried_metadata**: ['job', 'age', 'age_band', 'sign']
- **citation**: Schler, Koppel, Argamon, Pennebaker (2006). Effects of Age and Gender on Blogging. AAAI Spring Symposium.
- **license**: Non-commercial research use only (per the corpus authors).
- **gender_source**: self-reported by the blogger at registration

Gender is a binary proxy attribute used only to measure disparate model performance. It is not an endorsement of a binary model of gender, and for some sources it is inferred rather than self-reported (see provenance above).