Humanize Chinese AI Text v2.0

Comprehensive CLI for detecting and transforming Chinese AI-generated text. Makes robotic AI writing natural and human-like.

v2.0 highlights: weighted 0-100 scoring, sentence-level analysis, sentence restructuring (merge/split), context-aware replacement, rhythm variation, vocabulary diversification, 7 style transforms, external pattern config (patterns_cn.json).

Quick Start

# Detect AI patterns (20+ categories, 0-100 score)
python scripts/detect_cn.py text.txt
python scripts/detect_cn.py text.txt -v          # verbose + worst sentences
python scripts/detect_cn.py text.txt -s           # score only
python scripts/detect_cn.py text.txt -j           # JSON output

# Humanize text
python scripts/humanize_cn.py text.txt -o clean.txt
python scripts/humanize_cn.py text.txt --scene social
python scripts/humanize_cn.py text.txt --scene tech -a   # aggressive mode
python scripts/humanize_cn.py text.txt --seed 42         # reproducible

# Apply writing styles
python scripts/style_cn.py text.txt --style zhihu -o zhihu.txt
python scripts/style_cn.py text.txt --style xiaohongshu
python scripts/style_cn.py --list

# Compare before/after
python scripts/compare_cn.py text.txt --scene tech -a
python scripts/compare_cn.py text.txt -o clean.txt

Detection System

Scoring

Weighted 0-100 score with 4 severity levels:

Detection Categories

🔴 Critical (weight: 8)

🟠 High Signal (weight: 4)

🟡 Medium Signal (weight: 2)

⚪ Style Signal (weight: 1.5)

Sentence-Level Analysis

With -v (verbose) mode, the detector identifies the most AI-like sentences:

── 最可疑句子 ──
  1. [16分] 随着人工智能技术的不断发展，在当今数字化转型时代...
     原因: 数字化转型, 深度融合, 模板: 随着.*?的(不断)?发展

Humanization Engine

Transforms (applied in order)

1. Structure cleanup — Remove three-part structure (首先/其次/最后)
2. Phrase replacement — Context-aware replacement of AI phrases (regex patterns first, then plain text, longest-first matching)
3. Sentence merge — Merge overly short consecutive sentences
4. Sentence split — Split long sentences at natural breakpoints (但是/不过/同时)
5. Punctuation normalization — Reduce excessive semicolons, em dashes
6. Vocabulary diversification — Replace repeated words (进行/实现/提供 etc.) with synonyms
7. Paragraph rhythm — Vary uniform paragraph lengths (merge short, split long)
8. Casual injection — Add human expressions (scene-dependent)
9. Paragraph shortening — For social/chat scenes

Scenes

Aggressive Mode (-a)

Adds +0.3 casualness, more colloquial expressions, stronger sentence restructuring. Typical score reduction: 60-80 points on heavily AI-generated text.

Reproducibility

Use --seed N for reproducible results (same input + seed = same output).

Writing Style Transforms

7 specialized Chinese writing styles:

Combine humanize + style

python scripts/humanize_cn.py text.txt --style xiaohongshu -o xhs.txt

This first humanizes (removes AI patterns) then applies the style transform.

External Configuration

All patterns, replacements, and scoring weights are in scripts/patterns_cn.json. Edit this file to:

• Add new AI vocabulary patterns
• Customize replacement alternatives
• Adjust scoring weights per severity
• Add regex patterns for template detection
• Set thresholds for hedging language detection

Scripts Reference

detect_cn.py

python scripts/detect_cn.py [file] [-j] [-s] [-v] [--sentences N]

humanize_cn.py

python scripts/humanize_cn.py [file] [-o output] [--scene S] [--style S] [-a] [--seed N]

style_cn.py

python scripts/style_cn.py [file] --style S [-o output] [--seed N] [--list]

compare_cn.py

python scripts/compare_cn.py [file] [-o output] [--scene S] [--style S] [-a]

Shows score diff, category changes, and metric comparison before/after humanization.

Workflow

# 1. Check AI score
python scripts/detect_cn.py document.txt -v

# 2. Humanize with comparison
python scripts/compare_cn.py document.txt --scene tech -a -o clean.txt

# 3. Verify improvement
python scripts/detect_cn.py clean.txt -s

# 4. Optional: apply specific style
python scripts/style_cn.py clean.txt --style zhihu -o final.txt

Batch Processing

# Scan all files
for f in *.txt; do
  echo "=== $f ==="
  python scripts/detect_cn.py "$f" -s
done

# Transform all markdown
for f in *.md; do
  python scripts/humanize_cn.py "$f" --scene tech -a -o "${f%.md}_clean.md"
done