Paper Polishing and Duplication Reduction Toolkit

What It Addresses

Chinese academic papers often show AI-assisted writing traces, and naive rewriting can damage terminology, data, or argument flow. This skill targets concrete issues such as AI-likeness, high duplication, nonstandard terminology, and weak quality evaluation after editing.

How It Works

It combines rule-based checks with text-level features rather than relying on a single keyword list:

• AI-trace detection: ai_detector.py uses pattern matching, burstiness, TTR, perplexity, and information density to produce a 0–100 risk score and segment-level features.
• Four-layer gate: deai_gate.py fuses terminology, lexical, translation-style, and stylistic signals, focusing on AI-rewritten text that merely swaps synonyms.
• Rewriting and duplication reduction: segments scoring 35 or higher are edited with filler removal, structure variation, sentence-length variation, and protected-term-aware replacement.
• Terminology standardization: nonstandard terms are flagged and corrected using national terminology resources, MeSH 2026, and drug terminology.
• Quality report: before/after AI scores, terminology compliance, and readability indicators are summarized.

Boundaries

It is useful for medical, engineering, CS, and social-science paper editing, but the scores are probabilistic and not a final detection verdict. Data, citations, and statistics must remain unchanged; user confirmation is required before applying edits, and human review is still necessary for highly specialized or heavily formatted manuscripts.