SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsarxiv

arxiv

Writing & research

View repository

Updated 10-16 08:44

About this skill

Arxiv Paper Search

Execution Flow

用户请求 → 步骤1:确认需求 → 步骤2:构建查询 → 步骤3:执行搜索 → 步骤4:整理结果 → 步骤5:后续服务

Step 1: Confirm Requirements

Ask the user two questions:

Question 1: Depth of Organization

How would you like me to organize the search results?

  1. Quick Overview: Table overview + brief summary
  2. In-depth Analysis: Categorized organization + detailed explanation of key papers + research trends
  3. Literature Review: Save to file + complete academic review

Question 2: Number of Results

Scenario Suggested Number
Quickly understand a field 10-15 papers
In-depth study of a topic 20-30 papers
Writing a literature review 30-50 papers
Tracking the latest developments 5-10 papers

Step 2: Build the Search Query

Query Syntax

arxiv supports the following search syntax:

Syntax Meaning Example
AND Contains all LLM AND reasoning
OR Contains any GPT OR LLaMA
ANDNOT Exclude transformer ANDNOT vision
"..." Exact match "chain of thought"
ti: Title search ti:attention mechanism
au: Author search au:Hinton
abs: Abstract search abs:reinforcement learning
cat: Category search cat:cs.CL

Common Category Codes

Computer Science (cs)

  • cs.AI - Artificial Intelligence
  • cs.CL - Computational Linguistics (NLP)
  • cs.CV - Computer Vision
  • cs.LG - Machine Learning
  • cs.NE - Neural and Evolutionary Computing
  • cs.RO - Robotics
  • cs.SE - Software Engineering

Statistics (stat)

  • stat.ML - Machine Learning (statistics track)

Electrical Engineering (eess)

  • eess.AS - Audio and Speech Processing

Query Building Strategies

User says "find LLM-related papers" → Basic query:LLM OR "large language model" → If the results are too broad, ask for a specific direction and refine

User says "find papers by a certain author" → Use au:作者名 → Note: the arxiv author name format is lastname_firstname

User says "find the latest developments in a certain field" → Combine field keywords + restrict category → Example:"instruction tuning" AND cat:cs.CL

User gives a vague requirement → First use a broad query to get 5-10 papers → Optimize the query based on keywords in the results → Search again to get more precise results

Step 3: Execute the Search

Script location:scripts/search.py(relative to this skill directory)

python scripts/search.py "查询语句" -n 数量

Parameter Description：

  • First parameter: query statement (required)
  • -n: number of results to return (default 20)

Example：

# 搜索 LLM 推理相关论文，返回 30 篇
python scripts/search.py "LLM AND reasoning" -n 30

# 搜索特定作者
python scripts/search.py "au:bengio" -n 15

# 搜索特定分类下的主题
python scripts/search.py "attention AND cat:cs.CV" -n 20

Dependencies: requires installing the arxiv Python library (pip install arxiv）

Step 4: Organize the Results

Mode A: Quick Overview

Output structure：

## 搜索结果

[表格：序号 | 标题 | 作者 | 日期 | PDF]

## 快速总结

这批论文主要聚焦于 [领域]，核心方向包括：
1. [方向1]：[代表性工作]
2. [方向2]：[代表性工作]

值得优先关注的论文：#X、#Y、#Z

Mode B: In-depth Analysis

Output structure：

## 搜索结果

[表格]

## 领域概览

[1-2 段描述研究领域背景和当前主要方向]

## 主题分类

### 主题 1：[名称]（N 篇）

| 论文 | 核心贡献 | 方法特点 |
|------|----------|----------|
| #1 标题 | 解决了什么问题 | 用了什么方法 |
| #2 标题 | ... | ... |

### 主题 2：[名称]（N 篇）
...

## 关键论文详解

### 推荐 1：[论文标题]
- **问题**：该论文要解决什么问题
- **方法**：核心技术方案
- **结果**：主要实验结论和数据
- **价值**：为什么值得深入阅读

### 推荐 2：...

## 研究趋势

### 热点方向
- [方向1]：被 N 篇论文关注，代表性工作...
- [方向2]：...

### 新兴方法
- [方法1]：被 N 篇论文采用，特点是...
- [方法2]：...

### 开放问题
- [问题1]：多篇论文提到但尚未解决
- [问题2]：...

Mode C: Literature Review

Steps：

  1. Ask the user for the save path (default:./arxiv-review-{主题}-{日期}.md）

  2. Generate a review document, including:

    • Abstract (within 200 words)
    • Research background
    • Topic classification and detailed explanation (same as Mode B)
    • Research trend analysis
    • Research gaps and future directions
    • Reference list

  3. Reference format：

    [1] 作者. 标题. arXiv:XXXX.XXXXX, 年份. URL

Step 5: Follow-up service

After completing the organization, ask:

Do you need to read a specific paper in depth? Please tell me the number or title.

In-depth analysis of a single paper

If the user selects a paper, provide the following analysis:

## [论文标题]

**基本信息**
- 作者：[作者列表]
- 发布日期：[日期]
- PDF：[链接]
- 分类：[arxiv 分类]

**摘要翻译**
[中文翻译，保持学术准确性]

**核心内容**

### 1. 研究问题
- 该论文要解决什么问题？
- 为什么这个问题重要？
- 现有方法的局限性是什么？

### 2. 方法创新
- 核心技术方案
- 与现有方法的区别
- 关键设计选择及其理由

### 3. 实验结论
- 主要实验设置
- 核心结果数据
- 消融实验发现

### 4. 局限与展望
- 作者承认的局限
- 潜在的改进方向

**阅读建议**
- 如果你关注 [X]，重点看 Section [N]
- 如果你关注 [Y]，重点看 Section [M]

Paper quality assessment criteria

When recommending key papers, evaluate according to the following criteria:

Dimension High-quality signal Low-quality signal
Problem value Solves real pain points, has broad impact Marginal problem, lacks application scenarios
Methodological innovation New ideas, new frameworks, new paradigms Simple improvement, parameter tuning
Sufficient experiments Multiple datasets, ablation studies, fair comparison Single dataset, lacks comparison
Significant results Substantial improvement, breaks through bottlenecks Minor improvement, within the margin of error
Author background Well-known team, top conference record First publication, no background

Note: arXiv is a preprint and has not undergone peer review. For important decisions, it is recommended to track whether the paper has been accepted by a top conference.

Handling edge cases

Search results are empty

  1. Check whether the query syntax is correct
  2. Try broader keywords
  3. Remove category restrictions
  4. Inform the user and suggest alternative queries

Too many search results and they are scattered

  1. Ask the user for a more specific area of interest
  2. Add category restrictions (cat:cs.XX）
  3. Use more precise phrase search ("..."）

User needs are vague

Do not guess; proactively ask:

  • "Do you want to learn about [Direction A] or [Direction B]?"
  • "Are you more interested in theoretical methods or application scenarios?"
  • "Do you need to limit it to a certain time range?"

Non-English paper needs

arXiv mainly contains English papers. If the user needs Chinese-language literature:

  • Explain arXiv's language limitations
  • Suggest using Chinese academic platforms such as CNKI and Wanfang

Summary principles

  • Focus on core contributions: Explain what each paper did in 1-2 sentences
  • Avoid listing: Refine and synthesize, do not copy the abstract
  • Establish connections: Point out the relationships between papers (complementary, contrasting, improving, citing)
  • Highlight value: Help users judge which papers are worth investing time in
  • Stay objective: Present the limitations of the papers, do not overpraise