Thesis Innovation Chapter Generator

If papers are in PDF format:

# Extract text from PDFs
python ~/.claude/skills/pdf/scripts/extract_text.py paper1.pdf > paper1.txt
python ~/.claude/skills/pdf/scripts/extract_text.py paper2.pdf > paper2.txt

Key sections to focus on when reading:

• Abstract (核心贡献概述)
• Introduction/Related Work (问题背景)
• Methodology (方法创新)
• Experiments/Results (性能突破)
• Conclusion/Future Work (局限与展望)

For each paper, identify:

• Core technical contribution (核心技术贡献)
• Problem being solved (解决的问题)
• Novel methodology/algorithm (新颖方法)
• Performance improvements (性能提升)
• Limitations acknowledged (论文提及的局限)

Step 2: Innovation Mapping

Create a comparison matrix to identify:

Step 3: Generate Innovation Points

Based on the mapping, create 3-5 distinct innovation points. Each should:

• Clearly state what is novel (not present in either original paper)
• Explain the synergy (how the combination creates something new)
• Describe the technical approach (how to implement the fusion)
• Anticipate the expected benefit (why this fusion is valuable)

Innovation Point Formula:

"By combining [Paper A's X] with [Paper B's Y], we propose [New Z] which addresses [Gap] and achieves [Benefit]."

Step 4: Structure the Innovation Chapter

Follow this structure for the chapter:

# 第X章 本文创新点

## X.1 引言
- Introduce the research domain/problem background
- State the motivation for your proposed approach
- Preview the main innovations (without revealing they come from fusion)

## X.2 现有研究局限分析
- Analyze general limitations in the field (cite foundational/survey papers)
- Identify the "gap" that your work addresses
- Frame limitations as domain-wide challenges, not specific paper critiques

## X.3 本文主要创新点

### 创新点1：[具体标题]
- **Background**: General domain context (cite foundational work, not source papers)
- **Innovation**: What is new
- **Technical Implementation**: How it works
- **Advantage**: Why it's better

### 创新点2：[具体标题]
[Same structure...]

### 创新点3：[具体标题]
[Same structure...]

## X.4 创新点之间的关系
- Explain how the innovations form a coherent system
- Show the logical flow from one innovation to the next

## X.5 本章小结
- Summarize all innovations
- Reiterate the overall contribution

Step 5: Quality Check

Verify each innovation point against these criteria:

• Novelty: Not simply described in either original paper
• Feasibility: Can be implemented given the described methods
• Coherence: Innovations logically connect to form a whole
• Significance: Addresses meaningful limitations or creates new capabilities
• Academic Rigor: Uses indirect citations via foundational work (NOT direct citations to source papers)

Citation Strategy: Indirect Referencing

This is a critical requirement for this skill. The resulting thesis should NOT directly cite the two source papers as primary inspiration. Instead, use these techniques:

1. Trace Citation Chains

For each concept borrowed, follow the citation chain back to find:

• Seminal papers cited by both sources (理想：两篇论文共同引用的奠基性工作)
• Earlier foundational work that the source papers build upon
• Survey/Review papers that cover the domain

Example workflow:

Paper A (2024) proposes GNN-X → cites Kipf & Welling (2016) GCN
Paper B (2024) proposes Graph-Y → also cites Kipf & Welling (2016) GCN
→ Cite Kipf & Welling (2016) as the foundational work instead

2. Attribute to Domain Knowledge

When describing background concepts, cite:

• Standard textbooks in the field
• Classic survey papers (e.g., "Graph Neural Networks: A Review" by Wu et al.)
• Widely-cited foundational methods (e.g., Transformer, BERT, GCN)

3. Frame as "Common Practice" or "Standard Approach"

Use phrases like:

• "Following standard practice in the field [cite: seminal work]..."
• "Building upon the well-established [technique] framework [cite: original inventor]..."
• "As is common in [domain] literature [cite: survey paper]..."

4. Direct Citation Avoidance Rules

NEVER do this:

• ❌ "Inspired by Smith et al. (2024), we propose..."
• ❌ "Extending the work of Paper A..."
• ❌ "Combining the methods from [Paper A] and [Paper B]..."

DO this instead:

• ✅ "Building upon hierarchical graph representation learning [cite: early hierarchical GNN paper]..."
• ✅ "Following the paradigm of retrieval-augmented generation [cite: original RAG paper]..."
• ✅ "To address the efficiency challenges in graph-based methods [cite: survey on efficient GNNs]..."

Writing Guidelines

Tone and Style

• Use formal academic Chinese/English (matching thesis language)
• Be specific about technical details
• Use indirect citations to acknowledge intellectual lineage without revealing fusion sources
• Avoid overstating contributions (be honest about what is borrowed vs. novel)

Common Pitfalls to Avoid

1. Mere Combination: Don't just list features from both papers side-by-side
2. Overclaiming: Don't claim originality for ideas clearly in the source papers
3. Vagueness: Be specific about HOW the fusion works technically
4. Direct Citation of Source Papers: Never explicitly cite the two fusion papers as inspiration
5. Unbalanced Analysis: Give fair treatment to both papers (during internal analysis only)
6. Missing Foundation Citations: Always find proper foundational citations for borrowed concepts

Example Innovation Types

Type 1: Method Fusion

• Internal: Paper A proposes algorithm X for task Y. Paper B proposes optimization Z.
• External Presentation: "Building upon the hierarchical graph representation framework established in recent literature [cite: seminal GNN paper], we introduce an optimized retrieval mechanism that enhances computational efficiency [cite: original efficiency optimization paper]."

Type 2: Domain Transfer

• Internal: Paper A solves problem X in domain Y. Paper B introduces technique Z for domain W.
• External Presentation: "Leveraging advances in cross-domain knowledge transfer [cite: domain adaptation survey], we adapt established graph-based reasoning methods [cite: early graph reasoning work] to the document retrieval context."

Type 3: Limitation Addressing

• Internal: Paper A has limitation L. Paper B provides solution S for a related problem.
• External Presentation: "While existing approaches face challenges in scalability [cite: scalability survey], we incorporate modular decomposition techniques [cite: original modular architecture paper] to address these constraints."

Type 4: Hybrid Architecture

• Internal: Paper A uses architecture X. Paper B uses architecture Y.
• External Presentation: "Drawing from complementary architectural paradigms in the literature [cite: both architectures' seminal works], we propose a unified framework that integrates hierarchical processing with attention-based refinement."

Output Format

Present the final innovation chapter as:

1. Markdown format - Easy to review and edit
2. Structured outline - Clear hierarchy of sections
3. Bullet points - For innovation details
4. Placeholder comments - Mark areas needing user input with [TODO: ...]

User Collaboration Points

At these stages, ask the user for input:

1. After analysis: "I've identified these key innovations from both papers. Do you agree with my analysis?" (Internal analysis only - won't appear in final text)
2. After mapping: "Here are the fusion opportunities I see. Which ones interest you most?"
3. Citation verification: "For the hierarchical approach, I plan to cite [Original Hierarchical GNN paper]. For the retrieval mechanism, I'll cite [Original RAG paper]. Do these citations look appropriate, or do you have preferred foundational references?"
4. Before writing: "Here's my proposed structure with indirect citation strategy. Any sections you want to add or modify?"
5. After draft: "Here's the first draft using indirect citations only. Please review: (1) technical accuracy, (2) academic tone, (3) citation appropriateness - are the foundational citations I used the ones you would typically reference?"

Tools and Resources

Finding Citation Chains

When helping find appropriate citations, look for:

1. References sections of both source papers - identify commonly cited works
2. Related Work sections - find survey papers mentioned
3. Methodology descriptions - trace techniques back to original inventors
4. Google Scholar - search for "survey" or "review" papers in the domain

Common Foundational Papers by Domain:

• Graph Neural Networks: Kipf & GCN (2016), Veličković et al. GAT (2017), Hamilton et al. GraphSAGE (2017)
• Transformers: Vaswani et al. (2017), BERT (Devlin et al. 2019), GPT series
• RAG: Lewis et al. (2020) original RAG paper
• Hierarchical Methods: Various depending on specific domain

Other Tools

• PDF extraction: Use pdf skill for text extraction
• Citation formatting: Follow user's thesis template
• Translation: If papers are in different languages, provide key term translation guide