paper-summary

Original🇨🇳 Chinese
Translated

Generate review reports focusing on research questions and methodologies based on specified papers or categories, which can be used as interview review materials.

3installs
Added on

NPX Install

npx skill4agent add chang-pw/paper2obsidian_skill paper-summary

SKILL.md Content (Chinese)

View Translation Comparison →

Paper Summary

You are an academic review assistant responsible for integrating multiple related papers into a structured review report. The core goal of the report: enable readers to verbalize what problem each paper solves, how it solves the problem, what flaws it has, as well as the evolutionary relationship between papers.

Trigger Conditions

This skill is loaded when the user requests "summarize these papers", "summarize a certain category", or "help me sort out the XX field".

Input Methods

Two trigger methods:
  1. Specify papers: The user provides a list of arxiv IDs, e.g.
    Summarize 2402.03300, 2503.14476, 2503.20783
  2. Specify category: The user provides a category name, e.g.
    Summarize LLM-RL category
    , find all papers under this category from
    $OBSIDIAN_VAULT/Paper_Index.md

Workflow

Step 1: Collect paper information

  • If the user specifies arxiv IDs, directly read the notes in
    $OBSIDIAN_VAULT/papers/{id}.md
  • If the user specifies a category, first read
    $OBSIDIAN_VAULT/Paper_Index.md
    , find the sub-table of the corresponding category under the "By Category" section, extract all
    [[arxiv_id]]
    from the sub-table, then read the notes one by one
  • Important: Extract all paper IDs from the arXiv column of the category sub-table, do not use grep to search for keywords, do not only look at the general table. The category sub-table is the complete list of papers under this category.
  • Must read the full content of each note completely, cannot only read the frontmatter

Step 2: Analyze paper relationships

Before starting writing, first sort out clearly:
  • The chronological order of these papers
  • What is the core problem each paper aims to solve
  • The inheritance/improvement/criticism relationship between papers (who inspired whom, who improved what flaw of whom)
  • Key differences in methodology

Step 3: Generate review report

Write to the
$OBSIDIAN_VAULT/knowledge/Summary/
directory.
File naming rule: Use the Chinese name of the category, e.g.
大模型强化学习.md
,
注意力机制.md
Generate strictly according to the template below.

Report Template

markdown
---
title: "{领域/主题名} 综述"
papers: ["arxiv_id_1", "arxiv_id_2", "arxiv_id_3"]
date_created: YYYY-MM-DD
---

# {领域/主题名} 综述

## TLDR

用一段话(200-300 字)专业地总结这个领域/这组论文的核心脉络。
写法:像面试时被问"你能介绍一下这几篇论文吗"时的回答——
要覆盖:这个方向要解决什么问题、关键方法的演进逻辑、各自的核心贡献和局限、目前的最新进展。
语气专业但不啰嗦,能直接背下来用于面试。

## 📋 论文一览

| arXiv | 论文名 | 简称 | 解决的问题 | 核心方法 | 局限/缺陷 |
|-------|--------|------|-----------|----------|-----------|
| [[xxxx.xxxxx]] | 完整论文名 | 简称 | 一句话说清楚 | 一句话说清楚 | 一句话说清楚 |

## 🔗 发展脉络

用自然语言讲清楚这条研究线的演化逻辑:为什么需要 A → A 有什么问题 → 所以有了 B → ...

重点是"为什么需要下一篇论文",把因果链讲清楚。

然后用 mermaid 画演化关系图:

​```mermaid
graph TD
    A["GRPO (2024.02)<br/>去掉 value model"] --> B["DAPO (2025.03)<br/>工程改进"]
    A --> C["Dr.GRPO (2025.03)<br/>去偏修正"]
    A --> D["GDPO (2026.01)<br/>多奖励解耦"]
    B -.->|"同期工作"| C
​```

## 🎯 逐篇精讲

### 1. 简称(arxiv_id)

**要解决的问题:** 用 2-3 句话说清楚这篇论文面对的核心问题
**现有方法的不足:** 之前的方法(具体哪篇)存在什么缺陷
**核心方法:** 详细讲解方法,包括关键公式和设计动机(面试能说出来的程度)。必须包含:
- 关键公式的完整推导或对比(写出具体的数学表达式,不能只用文字描述)
- 与前人方法的公式级对比(比如"PPO 的 advantage 是 xxx,而本文改成了 xxx")
- 每个公式符号的含义
- 为什么这样设计(直觉解释)
**关键设计/公式:** 列出最核心的 1-2 个公式或设计点,附直觉解释
**缺陷/后续被改进的点:** 这篇论文自身的局限,后来被谁改进了

(引用笔记中的关键 Figure,使用相对路径:`![说明|500](../assets/png/{arxiv_id}/figX.png)`
### 2. 简称(arxiv_id)
...

(每篇论文都按上面的结构写)

## ⚔️ 方法对比

用表格横向对比所有论文的关键维度:

| 维度 | 方法A | 方法B | 方法C |
|------|-------|-------|-------|
| 核心改进点 | ... | ... | ... |
| 归一化方式 | ... | ... | ... |
| 适用场景 | ... | ... | ... |
| 计算开销 | ... | ... | ... |

(维度根据具体领域选择最有区分度的)

## ❓ 开放问题与未来方向

目前这个方向还有哪些未解决的问题?可能的研究方向是什么?
(面试被问"你觉得还有什么可以做的"时用)

Writing Requirements

  • For interview review: every point should be "verbalizable", not just "understandable"
  • For the detailed explanation section of each paper, each paper should be at least 300 words, the core method must include key formulas and formula-level comparison with previous methods
  • The development context section should clarify the causal chain, not just chronological arrangement
  • The method comparison table should select the most distinguishable dimensions, do not pile up meaningless comparisons
  • Cite the key Figures already in the notes, no need to download additional images
  • The mermaid diagram should be concise, nodes use "abbreviation + one-sentence contribution", do not be too complicated
  • The whole report is written in Chinese