Semantic Scholar Paper Metadata Skill

Search and retrieve research paper metadata using the Semantic Scholar Academic Graph API.

Base URL: https://api.semanticscholar.org/graph/v1 Auth: No key required for basic use. For higher rate limits, get a free key at https://www.semanticscholar.org/product/api and pass it as x-api-key header.

Choosing the Right Endpoint

Field Selection

All endpoints accept a fields parameter — a comma-separated list. Default response only includes paperId and title. Always request the fields you need.

Common fields:

paperId, corpusId, externalIds, title, abstract, year, venue,
publicationDate, authors, citationCount, referenceCount,
influentialCitationCount, isOpenAccess, openAccessPdf,
fieldsOfStudy, s2FieldsOfStudy, publicationTypes

Nested fields use dot notation:

authors.name, authors.affiliations,
citations.title, citations.year, citations.authors,
references.title, references.externalIds,
openAccessPdf.url

Endpoint Details & curl Examples

1. Keyword Search — /paper/search

Best for: topic discovery, finding relevant papers by concept.

bash
1curl -G "https://api.semanticscholar.org/graph/v1/paper/search" \
2  --data-urlencode "query=transformer attention mechanism" \
3  --data-urlencode "fields=title,year,authors,citationCount,abstract" \
4  --data-urlencode "limit=10"

Filters (all optional, combine freely):

bash
1--data-urlencode "year=2020-2024"               # year range
2--data-urlencode "venue=NeurIPS,ICML"           # specific venues
3--data-urlencode "fieldsOfStudy=Computer Science"
4--data-urlencode "minCitationCount=50"
5--data-urlencode "openAccessPdf="               # open access only (empty value = true)
6--data-urlencode "publicationDateOrYear=2023-01-01:2023-12-31"

Pagination: Response includes total, offset, next. Use offset=N to page through results (max 1,000 total).

2. Bulk Search — /paper/search/bulk

Best for: large result sets, boolean queries, sorted/filtered exports.

Supports AND/OR/NOT operators and returns up to 1,000 results per call with token-based pagination (no 1,000 total cap).

bash
1curl -G "https://api.semanticscholar.org/graph/v1/paper/search/bulk" \
2  --data-urlencode "query=large language models AND safety" \
3  --data-urlencode "fields=title,year,citationCount,authors" \
4  --data-urlencode "sort=citationCount:desc" \
5  --data-urlencode "limit=100"

Sort options: paperId, publicationDate, citationCount (append :asc or :desc).

Pagination: Response includes a token field. Pass it as &token=... on the next call.

3. Title Match — /paper/search/match

Best for: "I have this paper title, give me its metadata." Returns a single best match or 404. Includes a matchScore.

bash
1curl -G "https://api.semanticscholar.org/graph/v1/paper/search/match" \
2  --data-urlencode "query=Attention Is All You Need" \
3  --data-urlencode "fields=title,year,authors,citationCount,externalIds"

4. Paper by ID — /paper/{paper_id}

Best for: fetching full metadata when you already have an identifier.

Supported ID formats (prefix as shown):

bash
1# By arXiv ID
2curl "https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762?fields=title,abstract,year,authors,citationCount,referenceCount,isOpenAccess,openAccessPdf"
3
4# By DOI
5curl "https://api.semanticscholar.org/graph/v1/paper/DOI:10.1145/3292500.3330919?fields=title,year,venue,citationCount"

5. Batch Paper Lookup — POST /paper/batch

Best for: fetching metadata for a list of known paper IDs (up to 500).

bash
1curl -X POST "https://api.semanticscholar.org/graph/v1/paper/batch?fields=title,year,citationCount" \
2  -H "Content-Type: application/json" \
3  -d '{"ids": ["ARXIV:1706.03762", "ARXIV:2005.14165", "DOI:10.18653/v1/2020.acl-main.463"]}'

6. Citations & References

bash
1# Papers that CITE this paper (incoming)
2curl "https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762/citations?fields=title,year,authors,citationCount&limit=20"
3
4# Papers this paper CITES (outgoing)
5curl "https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762/references?fields=title,year,authors&limit=20"

Response wraps each item in a citingPaper or citedPaper key:

json
1{ "data": [{ "citingPaper": { "title": "...", "year": 2023 } }] }

7. Author Lookup

bash
1# Find an author's ID by name
2curl -G "https://api.semanticscholar.org/graph/v1/author/search" \
3  --data-urlencode "query=Yoshua Bengio" \
4  --data-urlencode "fields=name,affiliations,paperCount,citationCount,hIndex"
5
6# Get their papers
7curl "https://api.semanticscholar.org/graph/v1/author/{authorId}/papers?fields=title,year,citationCount&limit=20"

Rate Limits & API Key

• Without a key: ~100 requests/5 min (shared pool)
• With a free API key: ~1 request/sec sustained

To use an API key:

bash
1curl -H "x-api-key: YOUR_KEY" "https://api.semanticscholar.org/graph/v1/paper/..."

Store the key in an env var: export S2_API_KEY=... Then use: -H "x-api-key: $S2_API_KEY"

Parsing Results

Use jq to extract fields from responses:

bash
1# List titles and citation counts from a search
2curl -s -G "https://api.semanticscholar.org/graph/v1/paper/search" \
3  --data-urlencode "query=BERT language model" \
4  --data-urlencode "fields=title,year,citationCount" \
5  --data-urlencode "limit=5" \
6| jq '.data[] | "\(.citationCount)\t\(.year)\t\(.title)"'
7
8# Get all author names from a paper
9curl -s "https://api.semanticscholar.org/graph/v1/paper/ARXIV:1706.03762?fields=authors.name" \
10| jq '[.authors[].name]'
11
12# Extract open access PDF URL
13curl -s "https://api.semanticscholar.org/graph/v1/paper/ARXIV:2005.14165?fields=openAccessPdf" \
14| jq '.openAccessPdf.url'

Workflow Tips

• Start with /paper/search/match when the user provides a full title — it's faster than a keyword search.
• Use /paper/{id} when they give you a DOI, arXiv link, or PubMed ID — extract the ID and call directly.
• For "papers about X" questions, use /paper/search with minCitationCount and year filters to surface foundational work.
• For citation counts and h-index exploration, combine /author/search → /author/{id}/papers sorted by citationCount:desc.
• Always request only the fields you need — smaller responses, faster calls.
• When displaying results, format as a table with title, year, venue, and citation count for readability.