Reference Enricher

Matching & confidence

When you don't supply a DOI, the tool asks Crossref for several candidate records and scores each against what you pasted, then keeps the best. The composite score blends:

• Title overlap — shared significant words (weight .30)
• Full-reference similarity — character-level match of the whole string, robust to parsing slips (.22)
• First-five-words — fuzzy match of the title opening (.18)
• Author surnames — proportion of your authors found (.18)
• Year — exact = full credit, ±1 year = half (.12)
• Bonuses — journal, volume and page agreement add up to +.10

Confidence levels

• Verified — you pasted a DOI and Crossref resolved it. The pasted text is still compared; a mismatch is flagged.
• High — score ≥ 70%, or near-perfect title overlap.
• Medium — score ≥ 45%. Usually right; glance at it.
• Low — a candidate was found but signals are weak. Review before use.
• No match — Crossref returned nothing usable.

Abstract sources

The matched DOI is tried against four providers in order, stopping at the first that returns text:

1. Europe PMC — clean abstracts, strong for biomedical and health work.
2. Crossref — publisher-deposited abstracts (markup stripped).
3. OpenAlex — broad coverage, reconstructed from its word index.
4. Semantic Scholar — wide net, with a TL;DR fallback when no full abstract exists. It often refuses requests made from a browser without an API key; when that happens it is skipped for the rest of the run and the log says so.

The provider that supplied each abstract is labelled on every row.

Inputs & exports

Accepts pasted text, .txt (one reference per line), or .csv (the longest-text column is detected automatically). A DOI anywhere in a line is detected and trusted.

Results export as Copy table (TSV for spreadsheets), CSV (every score column included), or RIS (for Zotero / EndNote / Mendeley). Exports respect the current filters.

Privacy & rate limits

Everything runs in your browser — references go only to the public APIs above, never to a server of ours. The email and key are used for the request and not stored.

Requests are sequential and gently throttled to stay within free-tier limits, and a run can be stopped at any point without losing the rows already retrieved. Adding a contact email joins the Crossref and OpenAlex polite pools; a Semantic Scholar key both lifts its throttle and is usually what allows its requests to succeed from a browser at all.

Reading the results

Hover a score value to see the per-signal breakdown. Filter by confidence, search any field, or restrict to rows that carry an abstract. Always open and verify Low / No-match rows, and check any row flagged DOI may not match — that means a supplied DOI resolved but its record doesn't resemble your text.