Skip to content

Module: Abstract

Trigger: abstract, 摘要, abstract structure, 摘要结构, check abstract, polish abstract, abstract diagnosis, 润色摘要, abstract review

Commands

bash
uv run python -B scripts/analyze_abstract.py main.tex                       # thesis 骨架诊断(默认)
uv run python -B scripts/analyze_abstract.py main.tex --degree master       # 硕士字数阈值
uv run python -B scripts/analyze_abstract.py main.tex --bilingual           # + 中英摘要一致性
uv run python -B scripts/analyze_abstract.py main.tex --max-chars 1500      # 覆盖字数上界
uv run python -B scripts/analyze_abstract.py main.tex --model five --lang en --max-words 250
uv run python -B scripts/analyze_abstract.py main.tex --json

Details

Default --model thesis diagnoses the structural skeleton of a Chinese degree-thesis abstract: opening object-positioning sentence -> problem paragraph -> colon-ended overview sentence -> numbered work paragraphs -> optional closing paragraph. See the “Degree-Thesis Abstract Skeleton (thesis model)” section in ../writing/abstract-structure.md for the 13 T-* checks. Because this skill serves degree theses only, thesis is the default. --model five retains the conference-paper five-element model (Background/Objective/Methods/Results/Conclusion) as a fallback. The five-element model systematically produces false positives for doctoral abstracts, such as judging Results MISSING when no number is present even though a compliant doctoral abstract may close qualitatively.

Length thresholds align with the Yanshan rules constants in check_spec: --degree doctor (default) requires 900-1200 Chinese characters, and --degree master requires 500-650. An explicit --max-chars overrides the upper bound. Unit tests lock the constants in both locations together.

Bilingual abstract consistency (--bilingual): in thesis mode, additionally compare the English Abstract with the Chinese abstract. B-ORD (ordinal alignment), B-NUM (numeric-set consistency, Error), B-ENUM (numbered-item count), and B-LEN (English abstract absent or too short) are [Script]; B-SEM (sentence-level semantic correspondence) is an [LLM] lane. Tense/voice is not implemented here. The report footer routes to the deai module's English-abstract-gated tense check; deai traces do not flow into this module.

For Chinese thesis writing, also check whether the abstract, innovation/contribution claims, and conclusion form a three-way closure. See ../writing/thesis-writing-guide.md.

Thesis mode outputs each check code with level, evidence excerpt, and recommendation. --model five outputs PRESENT / VAGUE / MISSING for each element.

Skill-layer response:

  1. Format the diagnosis as a structured report with ✅ / ⚠️ / ❌ markers
  2. Provide specific revision suggestions for VAGUE or MISSING elements
  3. If the user requests polishing, generate a revised abstract with [REVISED: ...] annotations
  4. Never fabricate data or add claims not in the original

Thesis-specific closure:

  • Abstract: whether the research problem, method, results, and significance are complete.
  • Innovations/main contributions: whether they agree with the methods and results in the abstract.
  • Conclusion and outlook: whether they answer the contributions in the abstract and introduction and state limitation boundaries.

See also: abstract-structure.md for the degree-thesis abstract skeleton (thesis model) section (T-/B- checks) and the legacy five-element model with detection heuristics. See conclusion.md for conclusion-chapter content checks.

Released under the MIT License.