semantic-scholar-api

semantic-scholar-api

When to invoke

• User has a seed paper and wants its citation neighborhood
• Building a recommender that extends past keyword match
• Author disambiguation (two people with same name)
• Finding "influential" papers in a topic (citation-weighted)
• Getting canonical abstract/DOI when arXiv's metadata is incomplete

Two complementary APIs

API	Strength	Rate limit (no key)
Semantic Scholar api.semanticscholar.org/graph/v1/	Rich citation graph, influence scoring, AI-extracted tldr	1 req/sec
OpenAlex api.openalex.org/	Larger coverage (250M+ works), institution + funding data	10 req/sec with email in User-Agent

Use S2 for depth (citations, references, influence), OpenAlex for breadth (coverage, metadata richness).

Semantic Scholar — core endpoints

GET /graph/v1/paper/{id}
GET /graph/v1/paper/{id}/citations    # who cites this paper
GET /graph/v1/paper/{id}/references   # what this paper cites
GET /graph/v1/paper/search?query=...
GET /graph/v1/author/{id}/papers

Paper ID formats S2 accepts

• arXiv: arXiv:2604.13012 or ARXIV:2604.13012
• DOI: 10.1038/s41586-023-06792-0
• PubMed: PMID:34567890
• CorpusId: CorpusId:12345678
• S2 SHA: 649def34f8be52c8b66281af98ae884c09aef38b

Minimal citation walk

import requests, time
BASE = "https://api.semanticscholar.org/graph/v1"
HEADERS = {"User-Agent": "research-tool/1.0 ([email protected])"}

def paper(pid, fields="title,abstract,tldr,year,citationCount,authors"):
    r = requests.get(f"{BASE}/paper/{pid}", params={"fields": fields}, headers=HEADERS, timeout=15)
    r.raise_for_status()
    return r.json()

def citations(pid, limit=20, fields="title,year,citationCount,authors"):
    r = requests.get(
        f"{BASE}/paper/{pid}/citations",
        params={"fields": fields, "limit": limit},
        headers=HEADERS, timeout=15,
    )
    r.raise_for_status()
    return [c["citingPaper"] for c in r.json().get("data", [])]

def references(pid, limit=20, fields="title,year,citationCount"):
    r = requests.get(
        f"{BASE}/paper/{pid}/references",
        params={"fields": fields, "limit": limit},
        headers=HEADERS, timeout=15,
    )
    r.raise_for_status()
    return [c["citedPaper"] for c in r.json().get("data", [])]

# Example: walk 1 hop out from an arxiv paper
seed = paper("arXiv:2604.13032")
print(seed["title"], "→", seed["tldr"])
time.sleep(1)  # rate limit
for citer in citations("arXiv:2604.13032", limit=5):
    print(" cited by:", citer["title"])