Research Assistant

Research Assistant Skill

Search and explore academic research using the Semantic Scholar API. Access over 214 million papers, 2.49 billion citations, and 79 million authors across all fields of science.

When to Use

• User asks about scientific or academic papers
• User wants to find research on a topic
• User asks for citations or references
• User wants to understand the state of research in a field
• User asks "what does the research say about..."
• User wants to find papers by a specific author
• User needs to find papers related to another paper

Available Operations

1. Paper Search: Find papers by keywords, title, or abstract 2. Author Lookup: Find papers by a specific researcher 3. Citation Exploration: Find papers that cite or are cited by a paper 4. Paper Recommendations: Get papers similar to a given paper 5. Paper Details: Get full metadata including abstract, authors, venue, year

Instructions

When a user asks about research or academic papers:

Step 1: Understand the Query

Identify what the user wants:

• Topic search: "papers about transformer architectures"
• Author search: "papers by Geoffrey Hinton"
• Citation search: "what papers cite this one"
• Recommendations: "papers similar to Attention Is All You Need"

Step 2: Load API Reference

ALWAYS read references/api_reference.md before making API calls to ensure you use the correct endpoints and parameters.

Step 3: Construct the Search

IMPORTANT: Make ONE well-constructed search query. Do not make multiple API calls with variations of the same query. The API has rate limits, and multiple calls waste quota.

For topic searches, use the paper search endpoint:

https://api.semanticscholar.org/graph/v1/paper/search?query=YOUR_QUERY

Tips for effective single queries:

• Use specific, focused keywords (e.g., "transformer attention mechanism" not just "transformer")
• Include relevant filters like year, minCitationCount, or fieldsOfStudy to narrow results
• Request enough fields in one call: title,authors,year,citationCount,abstract,tldr
• If the first query returns no results, ask the user to refine their search rather than trying variations

Key parameters:

• query: Search terms (required)
• fields: Which data to return (see API reference)
• limit: Number of results (default 10, max 100)
• offset: For pagination

Step 4: Present Results

Format results clearly with:

• Paper title
• Authors (first author et al. if many)
• Year published
• Citation count
• Brief abstract or TLDR if available
• Link to paper (use paperId to construct URL)

Common City Coordinates (for reference, similar to weather skill pattern)

Not applicable - this skill uses text-based search queries.

Paper ID Formats

Semantic Scholar accepts multiple ID formats:

Resources

ALWAYS read these before making API calls:

• references/api_reference.md - Complete Semantic Scholar API documentation
• references/search_tips.md - Advanced search techniques and query optimization

Examples

Example 1: Topic Search

User asks: "Find papers about large language models from 2023"

1. Load API reference 2. Call: https://api.semanticscholar.org/graph/v1/paper/search?query=large+language+models&year=2023&fields=title,authors,year,citationCount,abstract&limit=10 3. Format results with titles, authors, years, and citation counts

Example 2: Author Search

User asks: "What has Yann LeCun published recently?"

1. First find author ID via author search 2. Then get author's papers via author endpoint 3. Filter/sort by year for recent work

Example 3: Citation Analysis

User asks: "What papers cite 'Attention Is All You Need'?"

1. Find paper ID for "Attention Is All You Need" 2. Use citations endpoint to get citing papers 3. Present most influential citations (by citation count)

Example 4: Research Overview

User asks: "What does the research say about sleep and memory?"

1. Search for papers on "sleep memory consolidation" 2. Get top cited papers to find influential work 3. Summarize key findings from abstracts 4. Provide references for further reading

Notes

• The API is free and does not require authentication for basic use
• Rate limit: 1000 requests/second shared among unauthenticated users
• For heavy use, request an API key for dedicated rate limits
• Always include the fields parameter to get useful data back
• Papers include TLDRs (AI-generated summaries) when available
• Citation counts are updated regularly but may lag slightly

Semantic Scholar API Reference

Semantic Scholar provides free access to a corpus of over 214 million academic papers with rich metadata, citations, and AI-generated summaries.

Base URL

https://api.semanticscholar.org/graph/v1

Authentication

No API key required for basic use. Unauthenticated requests share a pool of 1000 requests/second.

For higher rate limits, request an API key at: https://www.semanticscholar.org/product/api

Include key in header: x-api-key: YOUR_KEY

---

Paper Search Endpoint

GET /paper/search

Search for papers by keyword query.

Parameters

Example Request

GET https://api.semanticscholar.org/graph/v1/paper/search?query=machine+learning&fields=title,authors,year,citationCount,abstract,tldr&limit=10&year=2023-2024

Example Response

{
  "total": 1250000,
  "offset": 0,
  "data": [
    {
      "paperId": "649def34f8be52c8b66281af98ae884c09aef38b",
      "title": "Example Paper Title",
      "year": 2023,
      "citationCount": 150,
      "authors": [
        {"authorId": "123456", "name": "Jane Smith"},
        {"authorId": "789012", "name": "John Doe"}
      ],
      "abstract": "This paper presents...",
      "tldr": {
        "model": "tldr@v2.0",
        "text": "A concise AI-generated summary of the paper."
      }
    }
  ]
}

---

Paper Details Endpoint

GET /paper/{paper_id}

Get detailed information about a specific paper.

Paper ID Formats

Example Request

GET https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762?fields=title,authors,year,citationCount,abstract,tldr,references,citations

---

Paper Fields Reference

Use these in the fields parameter (comma-separated):

Basic Fields

Citation Fields

Author & Venue Fields

Access Fields

Classification Fields

---

Author Search Endpoint

GET /author/search

Search for authors by name.

Parameters

Example Request

GET https://api.semanticscholar.org/graph/v1/author/search?query=Geoffrey+Hinton&fields=name,paperCount,citationCount,hIndex

---

Author Details Endpoint

GET /author/{author_id}

Get details about a specific author.

Author Fields

Example Request

GET https://api.semanticscholar.org/graph/v1/author/1741101?fields=name,paperCount,citationCount,hIndex,papers.title,papers.year

---

Paper Citations Endpoint

GET /paper/{paper_id}/citations

Get papers that cite a specific paper.

Parameters

Example Request

GET https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762/citations?fields=title,authors,year,citationCount&limit=20

---

Paper References Endpoint

GET /paper/{paper_id}/references

Get papers referenced by a specific paper.

Example Request

GET https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762/references?fields=title,authors,year,citationCount&limit=20

---

Paper Recommendations Endpoint

GET /recommendations/v1/papers/forpaper/{paper_id}

Get papers similar to a given paper.

Parameters

Example Request

GET https://api.semanticscholar.org/recommendations/v1/papers/forpaper/649def34f8be52c8b66281af98ae884c09aef38b?fields=title,authors,year,citationCount&limit=10

---

Bulk Paper Lookup

POST /paper/batch

Get details for multiple papers at once.

Request Body

{
  "ids": [
    "ARXIV:1706.03762",
    "DOI:10.18653/v1/N18-3011",
    "CorpusId:215416146"
  ]
}

Example Request

POST https://api.semanticscholar.org/graph/v1/paper/batch?fields=title,year,citationCount

Body: {"ids": ["ARXIV:1706.03762", "ARXIV:2106.15928"]}

---

Fields of Study

Common values for fieldsOfStudy filter:

• Computer Science
• Medicine
• Biology
• Physics
• Chemistry
• Mathematics
• Engineering
• Psychology
• Economics
• Sociology
• Political Science
• Environmental Science
• Materials Science
• Business
• Art
• History
• Philosophy
• Linguistics
• Geography
• Education
• Law
• Agricultural and Food Sciences

---

Rate Limits

---

Error Responses

---

Constructing Paper URLs

To link to a paper on Semantic Scholar:

https://www.semanticscholar.org/paper/{paperId}

Example:

https://www.semanticscholar.org/paper/649def34f8be52c8b66281af98ae884c09aef38b

---

Tips for Best Results

1. Use specific queries: "transformer attention mechanism" beats "AI" 2. Always specify fields: Reduces response size and speeds up requests 3. Use year filters: Narrow down to recent or specific time periods 4. Check `tldr` field: AI summaries save time when available 5. Use `minCitationCount`: Filter out obscure papers 6. Paginate large results: Use offset for more than 100 results

Research Search Tips & Best Practices

This guide helps you construct effective searches and present research findings clearly.

Query Construction

Basic Search Strategies

Be Specific, Not Broad

Use Domain Terminology Academic papers use technical language. Match it:

Advanced Query Techniques

Phrase Searching Use quotes for exact phrases (when supported):

• "attention is all you need" - finds exact phrase
• transformer attention mechanism - finds papers with all terms

Combining Concepts Include multiple related terms:

• "large language models" OR "LLMs"
• BERT transformer NLP

Field-Specific Searches Target specific areas with fieldsOfStudy:

• Computer Science papers: fieldsOfStudy=Computer Science
• Medical research: fieldsOfStudy=Medicine
• Cross-disciplinary: search multiple fields

---

Filtering Strategies

By Time Period

By Impact

High-Impact Papers Use minCitationCount to find influential work:

• Seminal papers: minCitationCount=1000
• Established work: minCitationCount=100
• Notable papers: minCitationCount=50
• Recent quality: minCitationCount=10 (for papers < 2 years old)

Note: New papers have few citations. For recent work, lower the threshold or remove it.

By Publication Type

Pro tip: Literature Review papers are great for understanding a field quickly.

By Open Access

Set openAccessPdf=true to only return papers with free PDFs. Useful when the user needs to actually read the papers.

---

Presenting Results

Standard Result Format

For each paper, present:

**[Paper Title]**
Authors: [First Author] et al. ([Year])
Citations: [count] | Venue: [journal/conference]
TLDR: [AI summary if available]
Abstract: [First 2-3 sentences if no TLDR]
Link: [Semantic Scholar URL]

For Literature Overviews

When summarizing a research area:

1. Start with the most-cited papers (foundational work) 2. Include recent highly-cited papers (current directions) 3. Note any survey/review papers (comprehensive overviews) 4. Group by subtopic if results span multiple areas

Example Output Format

## Research on [Topic]: Key Papers

### Foundational Work
1. **Attention Is All You Need** (Vaswani et al., 2017)
   - Citations: 95,000+ | The paper that introduced Transformers
   - Key contribution: Self-attention mechanism replacing RNNs

### Recent Advances (2023-2024)
2. **[Recent Paper Title]** (Author et al., 2024)
   - Citations: 150 | Published in NeurIPS
   - Key finding: [Summary]

### Review Papers
3. **A Survey of Transformers** (Lin et al., 2022)
   - Comprehensive overview of transformer architectures
   - Good starting point for understanding the field

---

Common Research Tasks

Task: Find Seminal Papers in a Field

1. Search with broad field terms 2. Set minCitationCount=500 or higher 3. Sort mentally by citation count 4. Look for papers from 5+ years ago (time to accumulate citations)

Task: Find State-of-the-Art

1. Search specific topic + "state of the art" or technique name 2. Filter to recent years: year=2023-2024 3. Include top conferences/venues 4. Check citation count relative to age

Task: Literature Review

1. First, find any existing Review papers: publicationTypes=Review 2. Find foundational papers (high citations, older) 3. Find recent papers (last 2-3 years) 4. Get citations and references of key papers 5. Synthesize themes across papers

Task: Find Related Work

1. Get the paperId of the starting paper 2. Use /paper/{id}/references - what does it build on? 3. Use /paper/{id}/citations - what builds on it? 4. Use recommendations endpoint for similar papers

Task: Track a Researcher

1. Search author by name 2. Get their papers sorted by year 3. Identify their most-cited work 4. Note recent publications and trends

---

Handling Edge Cases

No Results Found

If search returns nothing: 1. Simplify the query (fewer terms) 2. Remove year filters 3. Try alternate terminology 4. Check spelling of technical terms 5. Search broader category

Too Many Results

If overwhelmed by results: 1. Add minCitationCount filter 2. Narrow year range 3. Add fieldsOfStudy filter 4. Use more specific terms 5. Focus on Review papers first

User Asks About Very Recent Work

Recent papers (< 1 year) have few citations:

• Don't use minCitationCount
• Look at author reputation instead
• Check venue quality (top conferences)
• Note the work is recent and citations pending

User Asks About Non-Academic Sources

Semantic Scholar covers academic papers only. For:

• Blog posts, news: Use web search
• Patents: Use Google Patents
• Technical reports: May be in S2 if archived on ArXiv

---

Quality Signals

Indicators of Important Papers

• High citation count relative to age
• Published in top venues (Nature, Science, NeurIPS, ICML, ACL, etc.)
• Authors from major institutions
• Has many influential citations (not just total citations)
• Referenced by other highly-cited papers

Indicators of Relevant Results

• Title matches query intent
• Abstract discusses the specific topic
• TLDR confirms relevance
• Fields of study align with user's interest
• Year matches user's timeframe needs

---

Top Venues by Field

Computer Science / AI

• NeurIPS, ICML, ICLR (machine learning)
• ACL, EMNLP, NAACL (NLP)
• CVPR, ICCV, ECCV (computer vision)
• Nature Machine Intelligence, JMLR

Medicine / Biology

• Nature, Science, Cell
• NEJM, Lancet, JAMA
• Nature Medicine, Nature Biotechnology

Physics

• Physical Review Letters
• Nature Physics
• Science

Knowing top venues helps prioritize results when citation counts are similar.