Cross-checking multiple data sources

When to use this workflow

No single academic search finds everything. Every source has its own coverage strengths.

• For biomedical papers, start with PubMed, then use Semantic Scholar to add citation networks and multidisciplinary results
• For computer science papers, try Semantic Scholar first, then supplement with Google Scholar
• For patents, use Google Patents first. For topics in languages other than English, supplement with Google Scholar or a currently available Chinese resource entry point

This guide explains how to choose and combine sources for different research tasks.

Source quick reference

Your need	First choice	Second choice	Not recommended
Computer science / Mathematics / Physics	Semantic Scholar	Google Scholar	PubMed
Biomedicine / Clinical research	PubMed	Semantic Scholar	Google Patents
Multidisciplinary topics	Semantic Scholar	Google Scholar	—
Reviews / Highly cited papers	Semantic Scholar (comprehensive citation counts)	Google Scholar	—
Latest preprints	Google Scholar (including arXiv)	Semantic Scholar	PubMed
Patents	Google Patents	—	—
Non-English literature	Google Scholar	—	Semantic Scholar / PubMed
Matching full-text content, beyond titles and abstracts	Full-text search	—	—

See Data sources for detailed introductions.

Three common strategies

1. Primary source + cross-checking source (recommended)

Use a primary source for the main workflow and a second source to look for gaps.

Computer science example:

• Search "graph neural network" in the primary source, Semantic Scholar → Collect 30 highly cited papers
• Search the same keywords in Google Scholar → Check whether its top 20 results include papers missing from SS
• The sources may add a few different results, especially from non-English journals or recently published material

Biomedical example:

• Primary source: PubMed → Search precisely with MeSH terms
• Cross-checking source: Semantic Scholar → Use citation networks to add preprints not indexed by PubMed

2. Metadata search + full-text search

Full-text search can find content discussed in a paper's body that does not appear in its title or abstract.

• Example: You want to find "papers that mention failure cases of batch normalization in their discussion of results"
• A Semantic Scholar search for the topic "batch normalization" will not surface that level of detail
• A full-text search for "batch normalization failure" can match passages discussing it in the paper's body

3. Use three sources for a review

To broaden coverage when writing a review:

1. Run MeSH keyword searches in PubMed
2. Search the same topic in Semantic Scholar and expand through citation networks
3. Use Google Scholar to add non-English literature and preprints
4. Merge results from all three sources and deduplicate by paperId / DOI
5. Add the screened papers to a project for later reading, manual verification, and reference in research sessions

Project papers do not automatically become a preferred candidate pool for the Citation annotation tool. After annotation, still check each citation against the statement it supports.

Deduplicate across sources

Each source uses its own IDs:

• Semantic Scholar uses paperId
• PubMed uses PMID
• Google Scholar uses cluster ID

The most stable cross-source identifier is the DOI. If two results have the same DOI, they refer to the same paper. Papers without a DOI, such as some preprints and short conference papers, require fuzzy title matching.

Check for cross-source duplicates when adding papers to a project. Do not assume that copies of the same paper from different sources will always merge automatically.

Source-specific pitfalls

Source	Pitfall
Semantic Scholar	Citation counts lag; citationCount may be low for the newest papers
PubMed	Limited coverage of computer science and engineering
Google Scholar	No official API; relies on a third-party proxy and may be temporarily unavailable
Google Patents	Primarily useful for patent research, not academic literature
Full-text search	Covers only papers whose full text has been indexed and may be less comprehensive than SS

A practical checklist

When researching a new topic, work through this list:

• Search SS with 3–5 core keywords
• Pick 1–2 of the most highly cited papers and examine their references and citations ("seed expansion")
• Supplement with five searches using the same keywords in PubMed (biomedicine) or Google Scholar (computer science)
• Add papers to a project and deduplicate them to form a candidate list for the current question
• Cross-check once with full-text search using key phrases from the field to look for major omissions

This process can reduce gaps caused by relying on a single source, but it cannot guarantee coverage of all relevant literature. Verify key conclusions against the original papers.