thesis-writer

Thesis-Writer: Intelligent Thesis Writing System

🎯 When to Use This Skill / When to Use This Skill

Trigger Conditions / Trigger Conditions:

When the user needs to write a master's degree thesis in the field of computer deep learning
When the user needs to complete thesis typesetting based on the [Your School] LaTeX template
When the user needs systematic and professional thesis writing guidance
When users need to write a master's thesis in computer science deep learning
When users need to complete thesis typesetting based on [Your University] LaTeX template
When users need systematic and professional thesis writing guidance

Applicable Scenarios / Applicable Scenarios:

✅ Full-process master's thesis writing / Full-process master's thesis writing
✅ Thesis structure design and outline creation / Thesis structure design and outline creation
✅ Academic literature review and comparison / Academic literature review and comparison
✅ Thesis polishing, AI detection removal, plagiarism reduction / Thesis polishing, AI detection removal, plagiarism reduction
✅ LaTeX formatting and typesetting optimization / LaTeX formatting and typesetting optimization

🛠️ Core Capabilities / Core Capabilities

Capability Matrix / Capability Matrix

Capability / Capability	Implementation / Implementation	Resource / Resource
Thesis outline generation	LLM + knowledge guidance	`references/outline_templates.md`
Chapter content writing	LLM + domain knowledge	`references/writing_guide.md`
Related work collection	Script assistance + LLM analysis	`scripts/paper_search.py` + `references/related_work_framework.md`
LaTeX typesetting	Script automation + template	`scripts/latex_formatter.py` + University template
AI trace removal and plagiarism reduction	LLM polishing + script detection	`scripts/ai_detector.py` + `references/paraphrasing_strategies.md`
Thesis review and revision	LLM + review standards	`references/review_checklist.md`

📋 Workflow / Workflow

Core Five-Phase Workflow / Five-Phase Workflow

graph TD
    A[阶段1: 需求分析与选题] --> B[阶段2: 大纲设计与确认]
    B --> C[阶段3: 章节撰写与填充]
    C --> D[阶段4: 润色优化与去重]
    D --> E[阶段5: 格式审查与定稿]
    
    B --> B1{用户确认大纲?}
    B1 -->|需修改| B
    B1 -->|通过| C
    
    C --> C1{章节循环}
    C1 --> C2[绪论]
    C2 --> C3[相关工作]
    C3 --> C4[方法论]
    C4 --> C5[实验与分析]
    C5 --> C6[总结与展望]
    
    D --> D1{质量检查}
    D1 -->|未达标| D
    D1 -->|达标| E

📝 Detailed Workflow / Detailed Workflow

Phase 1: Requirements Analysis & Topic Selection (Requirements Analysis & Topic Selection)

Objective: Clarify the thesis direction, research content, and innovation points

Interactive question checklist:

Research direction refinement

Q1: 你的研究是深度学习的哪个具体方向?
- 计算机视觉 (CV)
- 自然语言处理 (NLP)  
- 推荐系统 (RecSys)
- 图神经网络 (GNN)
- 多模态学习 (Multimodal)
- 其他: __________

Research question identification

Q2: 你要解决的核心问题是什么?
- 用1-2句话描述你的研究motivation
- 现有方法有什么不足?
- 你的方法如何改进?

Inventory of existing results

Q3: 你目前已经有哪些材料?
[ ] 已完成的实验结果
[ ] 参考的论文列表
[ ] 代码实现
[ ] 初步的方法设计
[ ] 其他: __________

Thesis type determination

Q4: 你的论文属于哪种类型?
- 提出新方法/新模型
- 改进现有方法
- 综合性对比研究
- 应用型研究

Time planning

Q5: 论文完成时间要求?
- 预期完成日期: __________
- 答辩时间: __________

Output:

📄 Generate thesis_requirements.md: Includes research direction, core questions, innovation points, and time planning
Provides a foundation for subsequent outline design

Phase 2: Outline Design & Confirmation (Outline Design & Confirmation)

Objective: Generate a thesis outline that complies with academic standards and school requirements

Execution steps:

Step 2.1: Identify the thesis structure type

First ask the user about the organization method of the thesis:

Q: 你的论文包含几个独立的研究工作?

A. 单一工作: 一个核心方法 + 对应实验
   → 使用"单工作结构": 第3章方法 + 第4章实验

B. 双工作: 两个相关但独立的完整工作
   → 使用"双工作结构": 第3章工作1(含方法+实验) + 第4章工作2(含方法+实验)

C. 多工作: 三个及以上独立工作
   → 使用"多工作结构": 每章一个完整工作

请选择或说明你的情况。

Step 2.2: Load the corresponding template

Read references/outline_templates.md
Select an appropriate template based on the structure type and thesis type

Step 2.3: Generate a draft outline

Structure A: Single-work structure (traditional structure)

Based on the [your school] master's thesis structure, generate an outline containing the following chapters:

# [论文题目]

## 摘要 (Abstract)
- 中文摘要 (200-300字)
- 英文摘要 (200-300词)
- 关键词 (3-5个)

## 第1章 绪论
### 1.1 研究背景及意义
### 1.2 国内外研究现状
  #### 1.2.1 [子方向1]研究现状
  #### 1.2.2 [子方向2]研究现状
### 1.3 本文主要工作
### 1.4 论文组织结构

## 第2章 相关技术与理论基础
### 2.1 [基础技术1]
### 2.2 [基础技术2]
### 2.3 本章小结

## 第3章 [核心方法章节名称]
### 3.1 问题定义
### 3.2 整体框架
### 3.3 [模块1]设计
### 3.4 [模块2]设计
### 3.5 本章小结

## 第4章 实验设计与结果分析
### 4.1 实验设置
  #### 4.1.1 数据集
  #### 4.1.2 评价指标
  #### 4.1.3 实验环境与参数
### 4.2 对比实验
### 4.3 消融实验
### 4.4 结果分析与讨论
### 4.5 本章小结

## 第5章 总结与展望
### 5.1 研究工作总结
### 5.2 未来工作展望

## 参考文献

## 致谢

## 攻读学位期间取得的研究成果

Structure B: Dual-work structure (common structure on CNKI)

Applicable scenario: The thesis contains two related but independent complete works, each with its own method design, experimental validation, and results analysis.

# [论文题目]

## 摘要 (Abstract)
- 中文摘要 (200-300字)
  - 背景 + 问题
  - 工作1简述 + 工作2简述
  - 实验结果 + 结论
- 英文摘要 (200-300词)
- 关键词 (3-5个)

## 第1章 绪论
### 1.1 研究背景及意义
### 1.2 国内外研究现状
  #### 1.2.1 [工作1相关]研究现状
  #### 1.2.2 [工作2相关]研究现状
  #### 1.2.3 现有方法的局限性
### 1.3 本文主要工作
  - 明确说明包含两个工作
  - 阐述两个工作的关系(递进/互补/并列)
### 1.4 论文组织结构

## 第2章 相关技术与理论基础
### 2.1 [两个工作共同的基础技术1]
### 2.2 [两个工作共同的基础技术2]
### 2.3 [评价指标与数据集]
### 2.4 本章小结

## 第3章 [工作1名称] (完整的独立工作)
### 3.1 引言
  - 工作1的具体问题和动机
  - 与工作2的关系说明
  
### 3.2 问题定义与分析
  - 形式化定义
  - 问题特点分析
  
### 3.3 [工作1方法名称]
  #### 3.3.1 整体框架
  #### 3.3.2 [核心模块1]设计
  #### 3.3.3 [核心模块2]设计
  #### 3.3.4 算法描述
  
### 3.4 实验设计与分析
  #### 3.4.1 实验设置
    - 数据集
    - 基线方法
    - 评价指标
    - 参数设置
  #### 3.4.2 对比实验结果
  #### 3.4.3 消融实验
  #### 3.4.4 可视化分析
  #### 3.4.5 结果讨论
  
### 3.5 本章小结
  - 总结工作1的贡献
  - 为工作2铺垫

## 第4章 [工作2名称] (完整的独立工作)
### 4.1 引言
  - 工作2的具体问题和动机
  - 基于工作1的改进/扩展/互补
  
### 4.2 问题定义与分析
  - 形式化定义
  - 与工作1的差异分析
  
### 4.3 [工作2方法名称]
  #### 4.3.1 整体框架
  #### 4.3.2 [核心模块1]设计
  #### 4.3.3 [核心模块2]设计
  #### 4.3.4 算法描述
  #### 4.3.5 (可选)与工作1的联合优化
  
### 4.4 实验设计与分析
  #### 4.4.1 实验设置
    - 数据集(可与工作1相同或不同)
    - 基线方法
    - 评价指标
    - 参数设置
  #### 4.4.2 对比实验结果
  #### 4.4.3 消融实验
  #### 4.4.4 可视化分析
  #### 4.4.5 与工作1的对比/联合效果
  #### 4.4.6 结果讨论
  
### 4.5 本章小结
  - 总结工作2的贡献
  - 总结两个工作的协同效果

## 第5章 总结与展望
### 5.1 研究工作总结
  - 分别总结工作1和工作2
  - 总结两个工作的整体贡献
### 5.2 未来工作展望
  - 工作1的改进方向
  - 工作2的扩展方向
  - 整体框架的优化方向

## 参考文献

## 致谢

## 攻读学位期间取得的研究成果

Key points of the dual-work structure:

Types of relationship between the two works:
- Progressive relationship: Work 2 improves upon the results of Work 1
- Complementary relationship: Work 1 and Work 2 solve the same problem from different perspectives
- Parallel relationship: Work 1 and Work 2 solve related but independent sub-problems
Chapter organization principles:
- Each work is a complete closed loop of "method + experiment"
- Chapter 3 and Chapter 4 are structurally symmetrical for easy comparison
- Explain the connection between the two works in the introduction and summary of each chapter
Word count allocation (total about 40,000-50,000 words):
- Chapter 1: 8,000-10,000 words
- Chapter 2: 6,000-8,000 words
- Chapter 3: 12,000-15,000 words (complete content of Work 1)
- Chapter 4: 12,000-15,000 words (complete content of Work 2)
- Chapter 5: 2,000-3,000 words

Step 2.3: Outline detailing

Add for each subsection:

Writing points: The core content that the section should include
Estimated word count: Suggested length allocation
Reference materials: Papers/books that need to be cited

Step 2.4: User confirmation and iteration

📋 生成的大纲已保存到: outline.md

请审阅大纲,并回答:
1. 章节结构是否合理?
2. 是否需要添加/删除某些章节?
3. 章节命名是否准确反映内容?

如需修改,请指出具体调整建议。确认无误后,我们将进入内容撰写阶段。

Loop: Modify the outline according to user feedback until it is confirmed and approved

Output:

📄 outline.md: The final confirmed detailed outline
📄 writing_plan.md: Chapter writing order and time allocation plan

Phase 3: Chapter Writing & Content Filling

Goal: Complete high-quality content writing chapter by chapter according to the outline

Suggested writing order:

Chapter 2 (Related technologies) → Establish the theoretical foundation
Chapter 3 (Core method) → Core technical contribution
Chapter 4 (Experimental analysis) → Validate the effectiveness of the method
Chapter 1 (Introduction) → Write after gaining a comprehensive understanding of the work
Chapter 5 (Summary and outlook) → Summarize at the end
Abstract → Distill after the full text is completed

3.1 Chapter 1: Writing the Introduction

Step 3.1.1: Research background and significance

Guiding principles (refer to references/writing_guide.md):

From macro to micro: field background → specific problem
The importance and urgency of the problem
The value of solving the problem

Example structure:

1. 领域大背景 (2-3段)
   - 深度学习在XX领域的发展
   - 当前应用现状

2. 具体问题引出 (2-3段)
   - 现有方法存在的局限
   - 问题的挑战性

3. 研究意义 (1-2段)
   - 理论意义
   - 应用价值

AI-assisted writing:

基于你的研究方向: [从阶段1获取]
我将为你生成研究背景初稿。

生成中... 

初稿已生成,请审阅:
[展示生成的内容]

请提供反馈:
- 是否准确反映你的研究?
- 哪些部分需要调整?
- 是否需要补充具体案例?

Step 3.1.2: Domestic and international research status

Research assistance:

Use scripts/paper_search.py to collect related papers:

python scripts/paper_search.py \
  --query "your research keywords" \
  --venue "NeurIPS,ICML,ICLR,CVPR,ACL" \
  --years "2020-2024" \
  --output related_papers.json

Classification and organization:

Load references/related_work_framework.md to obtain a literature review framework:

将相关工作分为3-4个子方向:

### 1.2.1 [子方向1: 如"基于注意力机制的方法"]
- 代表工作1 [引用]: 简述贡献
- 代表工作2 [引用]: 简述贡献
- 小结: 这类方法的优势与不足

### 1.2.2 [子方向2: 如"基于图神经网络的方法"]
...

### 1.2.3 [子方向3: 如"多模态融合方法"]
...

总结: 现有方法的共性问题,引出本文工作的必要性

Comparison table generation:

\begin{table}[htbp]
  \centering
  \caption{相关工作对比}
  \begin{tabular}{cccc}
    \toprule
    方法 & 核心技术 & 优势 & 局限 \\
    \midrule
    方法1 \cite{ref1} & XXX & ... & ... \\
    方法2 \cite{ref2} & XXX & ... & ... \\
    本文方法 & XXX & ... & - \\
    \bottomrule
  \end{tabular}
  \label{tab:related_work_comparison}
\end{table}

Step 3.1.3: Main work of this thesis

Writing focus:

Targetedness: Address the issues pointed out in Section 1.2
Innovativeness: Clearly state the innovation points (3-5)
Contributions: List specific contributions

Template:

针对上述问题,本文提出了[方法名称],主要工作包括:

(1) [创新点1]: 简述内容和意义

(2) [创新点2]: 简述内容和意义

(3) [创新点3]: 简述内容和意义

本文的主要贡献如下:
• 贡献1: ...
• 贡献2: ...
• 贡献3: ...

3.2 Chapter 2: Related Technologies and Theoretical Foundations

Objective: Introduce the preliminary knowledge needed for subsequent chapters

Principles:

Only introduce the necessary foundational knowledge; do not write it like a textbook
Each technology/concept must have a clear role (where it is used later in the text)

Content Organization:

### 2.1 [基础技术1: 如"深度神经网络"]
#### 2.1.1 基本原理
- 数学定义
- 核心公式

#### 2.1.2 典型架构
- CNN/RNN/Transformer等(按需选择)

### 2.2 [基础技术2: 如"注意力机制"]
#### 2.2.1 自注意力机制
#### 2.2.2 交叉注意力机制

### 2.3 [相关评价指标]
- 介绍实验章节会用到的评价指标

Formula Formatting Standards:

% 带编号公式
\begin{equation}
\text{Attention}(Q, K, V) = \text{softmax}\left(\frac{QK^T}{\sqrt{d_k}}\right)V
\label{eq:attention}
\end{equation}

% 多行公式
\begin{align}
h_t &= \sigma(W_h x_t + U_h h_{t-1} + b_h) \label{eq:hidden} \\
y_t &= \text{softmax}(W_y h_t + b_y) \label{eq:output}
\end{align}

3.3 Chapter 3: Core Method Chapter

This is the core of the thesis and requires the most detailed design

Step 3.3.1: Problem Definition

Formal Description:

\textbf{定义3.1 (问题定义):} 
给定输入数据集 $\mathcal{D} = \{(x_i, y_i)\}_{i=1}^{N}$, 
其中 $x_i \in \mathbb{R}^{d}$ 为特征向量, $y_i \in \mathcal{Y}$ 为标签,
本文旨在学习映射函数 $f: \mathbb{R}^{d} \rightarrow \mathcal{Y}$,
使得预测误差 $\mathcal{L}(f(x), y)$ 最小。

Step 3.3.2: Overall Framework

Architecture Diagram Drawing:

Use TikZ or other tools to draw the system architecture diagram:

\begin{figure}[htbp]
  \centering
  \includegraphics[width=0.9\textwidth]{figures/framework.pdf}
  \caption{[方法名称]整体框架}
  \label{fig:framework}
\end{figure}

Textual Description:

如图\ref{fig:framework}所示,本文提出的XXX方法主要包括三个模块:

1. 数据预处理模块: 负责...
2. 特征提取模块: 通过...
3. 预测模块: 最终...

各模块协同工作,完成从输入到输出的完整流程。

Step 3.3.3 & 3.3.4: Detailed Design of Core Modules

Each module should include:

Motivation: Why this module is needed
Design Idea: How to implement it
Mathematical Description: Formulated expression
Algorithm Pseudocode: Key algorithm

Algorithm Pseudocode Example:

\begin{algorithm}[h]
\caption{[算法名称]}
\label{alg:main}
\begin{algorithmic}[1]
\Require 输入数据 $X$, 参数 $\theta$
\Ensure 输出结果 $Y$

\State \textbf{初始化:} $h_0 = \mathbf{0}$
\For{$t=1$ to $T$}
  \State $h_t = f_{\theta}(x_t, h_{t-1})$  \Comment{特征提取}
  \State $\alpha_t = \text{Attention}(h_t, H)$  \Comment{注意力计算}
  \State $c_t = \sum_{i=1}^{t} \alpha_{t,i} h_i$  \Comment{上下文聚合}
\EndFor
\State $y = g_{\theta}(c_T)$  \Comment{输出预测}
\State \textbf{返回:} $y$
\end{algorithmic}
\end{algorithm}

Complexity Analysis:

时间复杂度: O(...)
空间复杂度: O(...)

3.4 Chapter 4: Experimental Design and Results Analysis

Step 3.4.1: Experimental Setup

Dataset Introduction:

#### 4.1.1 数据集
本文在以下X个公开数据集上进行实验:

1. **数据集1**: 
   - 来源: [引用]
   - 规模: XX个样本
   - 特点: ...
   - 划分: 训练/验证/测试 = 80%/10%/10%

2. **数据集2**: ...

Evaluation Metrics:

采用以下评价指标:

\textbf{准确率 (Accuracy):}
\begin{equation}
\text{Acc} = \frac{\text{正确预测数}}{\text{总样本数}}
\end{equation}

\textbf{F1分数:}
\begin{equation}
F1 = 2 \cdot \frac{\text{Precision} \cdot \text{Recall}}{\text{Precision} + \text{Recall}}
\end{equation}

Experimental Environment and Parameters:

#### 4.1.3 实验环境与参数
- 硬件: NVIDIA RTX 3090 (24GB)
- 软件: Python 3.8, PyTorch 2.0
- 超参数设置:
  - 学习率: 1e-4
  - Batch size: 32
  - Epoch: 100
  - 优化器: AdamW

Step 3.4.2: Comparative Experiments

Baseline Selection:

Select 3-5 representative SOTA methods
Explain the reasons for selection

Results Table:

\begin{table}[htbp]
  \centering
  \caption{不同方法在XX数据集上的性能对比}
  \begin{tabular}{lcccc}
    \toprule
    方法 & Acc(\%) & Precision(\%) & Recall(\%) & F1(\%) \\
    \midrule
    Baseline1 \cite{ref1} & 85.3 & 84.2 & 86.1 & 85.1 \\
    Baseline2 \cite{ref2} & 87.1 & 86.5 & 87.8 & 87.1 \\
    Baseline3 \cite{ref3} & 88.4 & 87.9 & 88.9 & 88.4 \\
    \textbf{Ours} & \textbf{91.2} & \textbf{90.5} & \textbf{91.8} & \textbf{91.1} \\
    \bottomrule
  \end{tabular}
  \label{tab:main_results}
\end{table}

Results Analysis:

从表\ref{tab:main_results}可以看出:
(1) 本文方法在所有指标上均优于基线方法
(2) 相比最佳基线Baseline3,本文方法在Acc上提升了2.8个百分点
(3) 这验证了[核心创新点]的有效性

Step 3.4.3: Ablation Experiments

Purpose: Verify the contribution of each module

\begin{table}[htbp]
  \centering
  \caption{消融实验结果}
  \begin{tabular}{lc}
    \toprule
    模型变体 & Acc(\%) \\
    \midrule
    完整模型 & \textbf{91.2} \\
    \quad - 移除模块A & 88.7 (-2.5) \\
    \quad - 移除模块B & 89.3 (-1.9) \\
    \quad - 移除模块A和B & 86.1 (-5.1) \\
    \bottomrule
  \end{tabular}
  \label{tab:ablation}
\end{table}

Analysis:

消融实验表明:
(1) 移除任一模块都会导致性能下降
(2) 模块A的贡献最大(下降2.5个百分点)
(3) 两个模块协同作用,带来5.1个百分点的提升

Step 3.4.4: Visualization Analysis

Attention visualization/feature visualization, etc.:

\begin{figure}[htbp]
  \centering
  \includegraphics[width=0.8\textwidth]{figures/attention_vis.pdf}
  \caption{注意力权重可视化}
  \label{fig:attention_vis}
\end{figure}

3.5 Chapter 5: Summary and Outlook

Step 3.5.1: Research Work Summary

Template:

本文针对[研究问题],提出了[方法名称]。
主要完成了以下工作:

(1) [工作1总结]
(2) [工作2总结]
(3) [工作3总结]

实验结果表明,[核心结论],验证了本文方法的有效性。

Step 3.5.2: Future Work Outlook

Directions:

Directions for method improvement
Possibilities for application expansion
Deepening of theoretical analysis

未来将在以下方向继续研究:
(1) [方向1]: 具体内容...
(2) [方向2]: 具体内容...

3.6 Abstract Writing

Write after the full text is completed

Chinese Abstract Structure (200-300 words):

[背景1-2句] + [问题1句] + [方法2-3句] + [实验结果1-2句] + [结论1句]

English Abstract:

Accurate translation of the Chinese abstract
Pay attention to academic English expression standards

Keyword Selection:

3-5
Include: domain terminology + method name + application scenario

General Standards for Chapter Writing

After completing each chapter:

Self-check list:

[ ] 逻辑连贯,无跳跃
[ ] 公式推导正确
[ ] 图表清晰,有caption和label
[ ] 引用格式正确
[ ] 段落间有过渡
[ ] 每节有小结

User review:

第X章初稿已完成,请审阅:

[展示章节内容摘要]

请确认:
1. 内容是否符合预期?
2. 技术细节是否准确?
3. 是否需要补充内容?

确认后将进入下一章节撰写。

Output (when Phase 3 is completed):

📄 LaTeX source files for each chapter
📄 references.bib: BibTeX file for references
📄 Generated chart and figure files

Phase 4: Polishing & Optimization

Objective: Improve the quality of the thesis, remove AI traces, and reduce the plagiarism rate

4.1 Content Polishing

Step 4.1.1: Language Polishing

Focus Points:

Academic Standards: Avoid colloquial expressions
Logical Coherence: Natural transitions between paragraphs
Accuracy of Expression: Avoid vague statements

Example Comparison:

❌ "这个方法很好" 
✅ "该方法在准确率和效率上均表现出色"

❌ "我们做了实验"
✅ "本文在X个数据集上进行了对比实验"

AI-Assisted Polishing:

逐段审阅内容,提供润色建议:

原文: [段落内容]
润色: [改进后内容]
理由: [修改原因]

请确认是否采纳。

Step 4.1.2: Logic Optimization

Checklist:

Clear logical chain between chapters
Problem-Method-Validation forms a closed loop
Innovation points correspond to experimental design
Conclusion is consistent with experimental results

4.2 Removing AI Traces

Objective: Make the paper more characteristic of human writing

Strategies (Reference references/paraphrasing_strategies.md):

Strategy 1: Rewrite high-risk expressions

Common AI expressions → Academic rewrites:

"值得注意的是" → "需要指出"
"显著提升" → "提高了X个百分点"
"深入分析" → "详细考察"

Strategy 2: Add domain-specific expressions

Incorporate professional terminology and expression conventions of this field:

通用: "模型性能更好"
专业: "模型在收敛速度和泛化能力上均有所提升"

Strategy 3: Enrich sentence structures

Avoid monotonous sentence patterns:

❌ 连续多句"本文...本文...本文..."
✅ 变换主语和句式: "为了...""考虑到...""基于..."

Use of AI Detection Tools:

python scripts/ai_detector.py --input thesis.tex --output report.json

The output report will flag high-risk passages for targeted revision.

4.3 Plagiarism Check and Reduction

Step 4.3.1: Plagiarism Rate Check

Tools:

Plagiarism detection system designated by the school
Self-check in advance: PaperPass/Weipu, etc.

Step 4.3.2: Plagiarism Reduction Strategies

Methods (Reference references/paraphrasing_strategies.md):

Synonym substitution:

原: "该方法能够有效提升模型性能"
改: "该方法可显著改善模型表现"

Sentence restructuring:

原: "深度学习在计算机视觉领域得到了广泛应用"
改: "计算机视觉领域广泛采用深度学习技术"

Active-passive conversion:

原: "本文提出了一种新方法"
改: "一种新方法在本文中被提出"

Expansion or compression:

原: "使用注意力机制"
扩展: "引入注意力机制以捕捉长距离依赖关系"

Formulaic expression:

原: "计算两个向量的相似度"
改: "计算相似度 $s = \text{cos}(v_1, v_2)$"

Auxiliary Tools:

python scripts/paraphrase_assistant.py \
  --input high_similarity_segments.txt \
  --output suggestions.txt

Manual Review: AI provides rewriting suggestions, and the user confirms whether to adopt them

4.4 Professional Review

Step 4.4.1: Automated Check

Run the check script:

python scripts/thesis_checker.py --input thesis.tex

Check Contents:

Consistency of citation format
Continuity of figure and table numbering
Correctness of equation numbering
Validity of cross-references between chapters
Spelling and grammar errors

Step 4.4.2: Review According to Evaluation Standards

Load references/review_checklist.md, and check item by item according to the standards:

Academic Standards:

Sufficient citations, no omission of key literature
Complete experimental setup, reproducible
Clear statement of innovation points

Writing Quality:

Fluent language, no grammatical errors
Clear logic, well-organized structure
Standard figures and tables, with adequate supporting explanations

Technical Depth:

Detailed method description, with formulas and algorithms
Sufficient experiments, including comparisons and ablation studies
In-depth analysis, not merely stating results

Output:

📄 polishing_report.md: Polishing and deduplication report
📄 Optimized LaTeX file

Phase 5: Format Review & Finalization

Objective: Ensure the paper fully complies with Hebei University template specifications

5.1 LaTeX Format Check

Step 5.1.1: Template Specification Comparison

Based on HBUthesis.cls and HBUthesis.tex templates:

Cover Information:

% 检查项
[ ] 学校代码: 10075
[ ] 学号: 正确填写
[ ] 论文题目: 中英文一致
[ ] 学位申请人姓名
[ ] 指导教师信息
[ ] 专业学位类别: 工程硕士
[ ] 专业学位领域: 软件工程
[ ] 院系名称
[ ] 答辩日期: 格式正确

Abstract Format:

% 中文摘要
\begin{cnabstract}
[摘要正文]
\par\noindent\heiti{关键字\ } \songti 关键字1\quad 关键字2\quad 关键字3
\end{cnabstract}

% 英文摘要
\begin{enabstract}
[Abstract content]
\par\noindent\textbf{Keywords\ } Keyword1\quad Keyword2\quad Keyword3
\end{enabstract}

Chapter Format:

\chapter{章节标题}  % 第一级
\section{节标题}     % 第二级
\subsection{小节标题} % 第三级

Figure and Table Format:

% 图片
\begin{figure}[htbp]
  \centering
  \includegraphics[width=0.8\textwidth]{figure.pdf}
  \caption{图片标题}
  \label{fig:label}
\end{figure}

% 表格 (三线表)
\begin{table}[htbp]
  \centering
  \caption{表格标题}
  \begin{tabular}{cccc}
    \toprule
    ... \\
    \midrule
    ... \\
    \bottomrule
  \end{tabular}
  \label{tab:label}
\end{table}

Equation Format:

% 单个公式
\begin{equation}
  formula
  \label{eq:label}
\end{equation}

% 多行公式
\begin{align}
  line1 \label{eq:label1} \\
  line2 \label{eq:label2}
\end{align}

References:

\bibliography{references}
% 确保 references.bib 格式正确

Step 5.1.2: Automated format check

python scripts/latex_formatter.py \
  --template HBUthesis \
  --input thesis.tex \
  --check-only \
  --output format_report.txt

Step 5.1.3: Format correction

Based on the check report, correct the format issues:

python scripts/latex_formatter.py \
  --template HBUthesis \
  --input thesis.tex \
  --auto-fix \
  --output thesis_formatted.tex

5.2 Compilation and output

Step 5.2.1: LaTeX compilation

# 完整编译流程
xelatex thesis.tex
bibtex thesis
xelatex thesis.tex
xelatex thesis.tex

Step 5.2.2: Check compilation output

PDF generated normally
No compilation errors
Table of contents generated correctly
References displayed normally
Figure and table numbering correct
Cross-references valid

Step 5.2.3: PDF quality check

Fonts embedded correctly
Image resolution sufficient
Page margins meet requirements
Headers and footers correct

5.3 Final review

Read the full text:

Print or read on a large screen
Check overall coherence
Look for missed errors

Blind review simulation:

按照评审标准,模拟盲审评分:

1. 选题与创新 (20分): __/20
2. 理论基础 (15分): __/15
3. 研究方法 (20分): __/20
4. 研究成果 (25分): __/25
5. 写作质量 (10分): __/10
6. 规范性 (10分): __/10

总分: __/100

不足之处: [列出]
改进建议: [列出]

5.4 Generate the final version

Output file list:

thesis_final/
├── thesis.pdf                 # 最终PDF
├── thesis.tex                # 主文件
├── chapters/                 # 章节文件
│   ├── chapter1.tex
│   ├── chapter2.tex
│   ├── ...
├── figures/                  # 图片文件
├── references.bib           # 参考文献
└── HBUthesis.cls            # 模板文件

Version record:

thesis_versions/
├── v1.0_initial_draft.pdf
├── v2.0_after_polishing.pdf
├── v3.0_final.pdf
└── version_log.md

Submission preparation:

Generate the originality statement and authorization letter
Prepare the Word version abstract (if needed)
Prepare the defense PPT outline

🔧 Tools & Scripts

`scripts/paper_search.py` - Paper search tool

Purpose: Search related papers from academic databases

Usage:

python scripts/paper_search.py \
  --query "deep learning attention mechanism" \
  --venue "NeurIPS,ICML,CVPR" \
  --years "2020-2024" \
  --max-results 50 \
  --output papers.json

Output format:

{
  "papers": [
    {
      "title": "...",
      "authors": ["...", "..."],
      "venue": "NeurIPS",
      "year": 2023,
      "abstract": "...",
      "url": "...",
      "bibtex": "..."
    }
  ]
}

`scripts/latex_formatter.py` - LaTeX formatting tool

Purpose: Check and correct LaTeX formatting

Usage:

# 仅检查
python scripts/latex_formatter.py --input thesis.tex --check-only

# 自动修正
python scripts/latex_formatter.py --input thesis.tex --auto-fix --output thesis_fixed.tex

`scripts/ai_detector.py` - AI trace detection tool

Purpose: Detect AI-generated traces in the paper

Usage:

python scripts/ai_detector.py --input thesis.tex --output ai_report.json

Output: Mark high-risk paragraphs and revision suggestions

`scripts/paraphrase_assistant.py` - Rewriting assistant

Purpose: Assist with paraphrasing to reduce similarity

Usage:

python scripts/paraphrase_assistant.py \
  --input high_similarity.txt \
  --style academic \
  --output suggestions.txt

`scripts/thesis_checker.py` - Paper completeness check

Purpose: Comprehensively check paper quality

Usage:

python scripts/thesis_checker.py --input thesis.tex --output check_report.md

Check items:

Structural completeness
Citation consistency
Figure and table numbering
Spelling and grammar
Formatting standards

📚 References

`references/outline_templates.md`

Content: Outline templates for different types of papers

Method-innovation paper outline
Review paper outline
Applied paper outline

When to load: When generating the outline in Stage 2

`references/writing_guide.md`

Content: Academic writing standards and techniques

Writing focus for each chapter
Academic language standards
Common writing errors

When to load: When writing chapters in Stage 3

`references/related_work_framework.md`

Content: Framework for the related work review

Literature classification methods
Comparison table templates
Review writing techniques

When to load: When writing Section 1.2

`references/latex_guide.md`

Content: LaTeX usage guide

Instructions for using the Hebei University template
Common LaTeX commands
Figure, table, and formula typesetting techniques

When to load: During format adjustment in Stage 5

`references/paraphrasing_strategies.md`

Content: Rewriting and Similarity Reduction Strategy

Synonym Lexicon
Sentence Restructuring Methods
Academic Expression Improvement

When to Load: During Stage 4 similarity reduction optimization

`references/review_checklist.md`

Content: Thesis Review Standards

Academic Standards Checklist
Technical Depth Assessment
Writing Quality Standards

When to Load: During Stage 4 and Stage 5 review

💡 Usage Example / Usage Example

Complete Workflow Example

User:

我需要写一篇关于多模态情感分析的硕士论文,
使用Transformer和图神经网络结合的方法。

AI (Start thesis-writer):

Stage 1: Requirements Analysis

让我们开始梳理你的论文需求。

Q1: 研究方向细化
你的研究是: 
☑ 多模态学习 (Multimodal) - 情感分析方向

Q2: 核心问题
请用1-2句话描述:

After user responds...

Stage 2: Generate Outline

基于你的研究方向,我生成了初步大纲:

# 基于Transformer和图神经网络的多模态情感分析

## 第1章 绪论
### 1.1 研究背景及意义
### 1.2 国内外研究现状
  #### 1.2.1 基于Transformer的情感分析方法
  #### 1.2.2 基于图神经网络的多模态融合
  #### 1.2.3 多模态情感分析研究现状
### 1.3 本文主要工作
### 1.4 论文组织结构

...

[完整大纲]

请审阅并确认,或提出修改建议。

After confirmation, proceed to Stage 3 writing...

⚙️ Token Optimization Strategy / Token Optimization

Progressive Loading Principle:

Stages 1-2: Load only the outline template (~800 tokens)
Stage 3: Load chapter writing guides on demand (~1000 tokens/chapter)
Stage 4: Load polishing and similarity reduction strategies (~1200 tokens)
Stage 5: Load format checking standards (~600 tokens)

Avoid Redundancy:

Do not load all references simultaneously
Archive completed chapter content, do not reload
Use scripts for deterministic tasks to save LLM tokens

⚠️ Limitations & Notes / Limitations & Notes

Known Limitations:

Technical Content Accuracy:
- AI-generated technical details require user verification
- Domain-specific formula derivations require manual review
Novelty Assessment:
- AI cannot determine whether an innovation has already been published
- User must supplement with the latest literature review
Experimental Results:
- AI cannot generate real experimental data
- User must provide experimental results; AI assists with analysis and writing
Plagiarism Detection System Differences:
- Results may differ across plagiarism detection systems
- The school-designated system shall prevail

Best Practices:

✅ Phased Confirmation: Confirm with the user after each stage is completed
✅ User-Led: Technical content is vetted by the user, AI assists with expression
✅ Iterative Optimization: Allow multiple rounds of revision and refinement
✅ Version Management: Keep all versions for easy traceback

📦 Dependencies / Dependencies

Python Dependencies:

requirements.txt:

# 论文搜索
scholarly>=1.7.0
arxiv>=1.4.0

# LaTeX处理
pylatexenc>=2.10

# 文本处理
nltk>=3.8
jieba>=0.42

# 其他工具
requests>=2.28.0
beautifulsoup4>=4.11.0

LaTeX Dependencies:

XeLaTeX compiler
Hebei University thesis template (HBUthesis.cls)
Chinese fonts: SimSun, SimHei, FangSong, KaiTi

🗂️ File Structure / File Structure

thesis-writer/
├── SKILL.md                          # 本文件
├── requirements.txt                  # Python依赖
├── scripts/                          # 工具脚本
│   ├── paper_search.py              # 论文搜索
│   ├── latex_formatter.py           # LaTeX格式化
│   ├── ai_detector.py               # AI痕迹检测
│   ├── paraphrase_assistant.py      # 改写助手
│   └── thesis_checker.py            # 论文检查
├── references/                       # 参考文档
│   ├── outline_templates.md         # 大纲模板
│   ├── writing_guide.md             # 写作指南
│   ├── related_work_framework.md    # 综述框架
│   ├── latex_guide.md               # LaTeX指南
│   ├── paraphrasing_strategies.md   # 改写策略
│   └── review_checklist.md          # 评审标准
└── templates/                        # 模板文件
    └── HBUthesis/                   # 河北大学模板
        ├── HBUthesis.cls
        ├── HBUthesis.tex
        └── ...

Get Started: Tell me your research direction and thesis requirements, and let's complete a high-quality degree thesis together!

Get Started: Tell me your research direction and thesis requirements, let's create a high-quality thesis together!

About this skill

🎯 When to Use This Skill / When to Use This Skill

🛠️ Core Capabilities / Core Capabilities

Capability Matrix / Capability Matrix

📋 Workflow / Workflow

Core Five-Phase Workflow / Five-Phase Workflow

📝 Detailed Workflow / Detailed Workflow

Phase 1: Requirements Analysis & Topic Selection (Requirements Analysis & Topic Selection)

Phase 2: Outline Design & Confirmation (Outline Design & Confirmation)

Structure A: Single-work structure (traditional structure)

Structure B: Dual-work structure (common structure on CNKI)

Phase 3: Chapter Writing & Content Filling

3.1 Chapter 1: Writing the Introduction

3.2 Chapter 2: Related Technologies and Theoretical Foundations

3.3 Chapter 3: Core Method Chapter

3.4 Chapter 4: Experimental Design and Results Analysis

3.5 Chapter 5: Summary and Outlook

3.6 Abstract Writing

General Standards for Chapter Writing

Phase 4: Polishing & Optimization

4.1 Content Polishing

4.2 Removing AI Traces

4.3 Plagiarism Check and Reduction

4.4 Professional Review

Phase 5: Format Review & Finalization

5.1 LaTeX Format Check

5.2 Compilation and output

5.3 Final review

5.4 Generate the final version

🔧 Tools & Scripts

scripts/paper_search.py - Paper search tool

scripts/latex_formatter.py - LaTeX formatting tool

scripts/ai_detector.py - AI trace detection tool

scripts/paraphrase_assistant.py - Rewriting assistant

scripts/thesis_checker.py - Paper completeness check

📚 References

references/outline_templates.md

references/writing_guide.md

references/related_work_framework.md

references/latex_guide.md

references/paraphrasing_strategies.md

references/review_checklist.md

💡 Usage Example / Usage Example

Complete Workflow Example

⚙️ Token Optimization Strategy / Token Optimization

⚠️ Limitations & Notes / Limitations & Notes

📦 Dependencies / Dependencies

🗂️ File Structure / File Structure

`scripts/paper_search.py` - Paper search tool

`scripts/latex_formatter.py` - LaTeX formatting tool

`scripts/ai_detector.py` - AI trace detection tool

`scripts/paraphrase_assistant.py` - Rewriting assistant

`scripts/thesis_checker.py` - Paper completeness check

`references/outline_templates.md`

`references/writing_guide.md`

`references/related_work_framework.md`

`references/latex_guide.md`

`references/paraphrasing_strategies.md`

`references/review_checklist.md`