Citegeist: BibTeX Reference Checker

Checks, standardizes, and upgrades .bib files automatically.

Features

Feature	Description
Venue standardization	Detects common spellings of conference/journal names and replaces them with canonical @String macros
arXiv → published upgrade	Searches Semantic Scholar, CrossRef, arXiv, Google Scholar, and Perplexity AI to find the formal publication venue for preprints
DuckDuckGo verification	Confirms every found publication exists on the web to guard against LLM hallucinations
Entry type inference	Fixes @misc → @article or @inproceedings based on available fields
Missing field detection	Warns about required fields absent from entries
Duplicate key detection	Errors on duplicate cite keys
Undefined @String detection	Errors on bare-word macro references not defined anywhere

Installation

Requires Python ≥ 3.11 and uv.

uv sync

This installs the bib-check command into a local .venv.

Usage

uv run bib-check references.bib

Options:

Flag	Description
-o FILE	Output .bib file (default: <input>_fixed.bib)
-r FILE	Output report file in Markdown (default: <input>_report.md)
--offline	Skip all network lookups
--no-upgrade	Skip arXiv → published upgrade (venue standardization still runs)
--perplexity-key KEY	Perplexity AI API key (overrides PERPLEXITY_API_KEY env var)
--s2-key KEY	Semantic Scholar API key for higher rate limits (overrides SEMANTIC_SCHOLAR_API_KEY env var)
--no-scholar	Disable the Google Scholar search backend
--no-learn-venues	Do not save newly discovered venues to venues.json
-v	Verbose: print search progress to stderr

Semantic Scholar API key

Semantic Scholar is the primary search backend. By default requests are unauthenticated (~1 req / 3.5 s). Providing an API key raises the limit to ~10 req / s:

1. Get a free key at https://www.semanticscholar.org/product/api#api-key
2. Pass it via environment variable or CLI flag:

export SEMANTIC_SCHOLAR_API_KEY=...
uv run bib-check references.bib

# or inline:
uv run bib-check references.bib --s2-key ...

Perplexity AI integration

Perplexity AI has web-search capabilities and can find publication venues that structured APIs (Semantic Scholar, CrossRef) may not index yet.

1. Get a key at https://www.perplexity.ai/settings/api
2. Pass it via environment variable or CLI flag:

export PERPLEXITY_API_KEY=pplx-...
uv run bib-check references.bib

# or inline:
uv run bib-check references.bib --perplexity-key pplx-...

Every result from Perplexity is verified via DuckDuckGo before being accepted, so hallucinated DOIs / venues are discarded automatically.

Search pipeline

For each arXiv preprint the tool runs the following pipeline (in order, stopping at the first confirmed published result):

1. Semantic Scholar  (structured API, most complete venue data)
2. CrossRef          (DOI-indexed published works)
3. arXiv             (journal_ref / DOI sometimes present)
4. Google Scholar    (broad academic coverage; requires `scholarly` package)
5. Perplexity AI     (web search + LLM; requires API key)
   └→ DuckDuckGo     (verifies result to catch hallucinations)
6. DuckDuckGo        (soft-verify S2 / CrossRef results too)

Papers that cannot be confirmed as published are flagged for manual review in the report.

Google Scholar support

Google Scholar is used as a fallback search backend via the scholarly package.

uv sync --group scholar

Once installed, Google Scholar is enabled automatically. Disable it with --no-scholar if needed.

Output

• references_fixed.bib – cleaned bib with canonical @String macros and upgraded entries
• references_report.md – Markdown report with all changes and issues

@String macros emitted

Only macros that are actually used in the output file are emitted. The canonical set includes: CVPR, ICCV, ECCV, NeurIPS, ICML, ICLR, AAAI, IJCAI, AISTATS, SIGGRAPH, TOG, PAMI, IJCV, TIP, TVCG, TMM, TCSVT, WACV, ACMMM, BMVC, ICPR, CGF, EGSR, ARXIV, and more.

Project structure

/
├── pyproject.toml
└── bib_checker/
    ├── cli.py       – argument parsing and entry point
    ├── checker.py   – main checking logic
    ├── parser.py    – BibTeX parser (no external deps)
    ├── writer.py    – BibTeX serializer
    ├── search.py    – arXiv / S2 / CrossRef / Google Scholar / Perplexity / DDG backends
    ├── strings.py   – canonical @String definitions and alias table
    ├── datatypes.py – shared data classes
    └── report.py    – Markdown report generator