Added
- 13 AI-detection metrics — new perplexity_score metric (character-level trigram model)
- Ensemble boosting — 3-classifier aggregation: base weighted sum (50%), strong-signal detector (30%), majority voting (20%). 90.9% accuracy
- Benchmark suite — 11 labeled samples, per-label accuracy breakdown
- CLI detect subcommand —
texthumanize detect [file]with emoji verdicts - Streaming progress callback —
humanize_batch(texts, on_progress=callback) - C2PA / IPTC watermark detection — content provenance pattern detection
- Tone replacements for UK/DE/FR/ES — informal ↔ formal replacement pairs
- PHP examples/ — basic_usage.php & advanced.php
Changed
- Zipf metric rewritten — log-log linear regression with R² goodness-of-fit
- Confidence formula — 4-component with text length, metric agreement, extreme bonus
- Grammar detection expanded — 5 → 9 indicators
Full Changelog: https://github.com/ksanyok/TextHumanize/blob/main/CHANGELOG.md
Install
pip install git+https://github.com/ksanyok/TextHumanize.git@v0.7.0