How Reverse Lookup Works#

How Reverse Lookup Works#

Reverse lookup turns incomplete citation text into a ranked list of works. The normal request path searches OokCite’s local citation corpus; external services are bounded fallback sources.

1. Normalize the input#

OokCite trims whitespace, collapses line breaks, removes common reference numbers such as [32] or 1., preserves embedded DOI values, and removes URL or access-date boilerplate. Cleaned search text is capped at 500 UTF-8-safe bytes.

2. Plan the query#

The resolver separates likely author, title, year, journal, identifier, ORCID, and affiliation evidence. Identifier-shaped input uses exact lookup. A person-only query uses the structured author and ORCID paths so author tokens from different people are not treated as one identity.

3. Generate candidates#

Configured local candidate sources can include:

  • exact DOI, ISBN, and title indexes;

  • full-corpus lexical search;

  • structured author and alias projections;

  • the live in-process overlay for newly ingested metadata;

  • bounded semantic candidates when the local vector projection is available.

Candidate sources are merged by canonical identifier before final scoring. A local miss can trigger bounded provider lookups such as Crossref, DataCite, OpenLibrary, or a source-specific adapter. Provider metadata can be projected into the canonical corpus for later local lookup.

4. Rank bibliographic fit#

The ranker compares title, author identity and coverage, publication year, journal, identifier, source provenance, and available impact evidence. Exact identity constrains author results; citation count or author prestige cannot substitute for a bibliographic or person match.

OokCite returns ranked candidates when the evidence is ambiguous. The web client can present those alternatives instead of silently treating the first candidate as certain.

5. Cache and format#

Eligible reverse-query results are cached in memory for one hour. A reverse cache hit returns before lookup metering. Person-name and bare-ORCID queries skip that cache path so identity and ranking updates take effect immediately.

Formatting already-resolved metadata in another citation style is local work and does not consume lookup quota.

Limitations#

  • Short or ambiguous input may match several legitimate works.

  • Books without an ISBN or DOI can be confused with reviews or related editions.

  • Sparse author names can remain ambiguous without ORCID, affiliation, co-author, title, or year evidence.

  • Results depend on the active local projections and the metadata exposed by upstream sources.