Rebiber

Rebiber

Replace unofficial / arXiv BibTeX entries with official DBLP or ACL Anthology records. Cite keys are kept.

Web demo · Colab

Install from GitHub only. We do not publish or support PyPI (pip install rebiber is a 2021 1.1.3 release).

Install

Requires Python 3.10 or newer. Rebiber 1.4 uses bibtexparser 2 (>=2,<3).

uv tool install git+https://github.com/yuchenlin/rebiber
# or
pip install "rebiber @ git+https://github.com/yuchenlin/rebiber"

Develop from a checkout:

git clone https://github.com/yuchenlin/rebiber.git && cd rebiber
uv sync --extra dev          # or: pip install -e ".[dev]"
uv run pytest

Changes in 1.4

• Migrate to bibtexparser 2 and raise the minimum Python version from 3.8 to 3.10.
• Preserve single-space field indentation and one blank line between entries.
• Apply field removal, venue abbreviation, field selection, title protection, and sorting to entries with repeated cite keys when -d False is set.

Usage

Omitting -o overwrites each -i file. Start with --dry-run.

rebiber -i refs.bib --dry-run
rebiber -i refs.bib --dry-run --report changes.txt
rebiber -i refs.bib -o refs.official.bib
rebiber -i refs.bib --used-in paper.tex appendix.tex
rebiber -i refs.bib --live-lookup          # opt-in DBLP search after a local miss
rebiber -i *.bib -o ./normalized/
rebiber -i refs.bib --format-only -o pretty.bib
rebiber -i refs.bib --keep author,title,booktitle,journal,year,volume,number,pages,doi --protect-titles

Examples: examples/input.bib → examples/output.bib.

Flag	Default	Meaning
-i	required	One or more .bib files
-o	in place	Output file, or a directory if -i has multiple files
--dry-run	off	Print the change report; do not write .bib
--report PATH	—	Also write the change report to a file
--used-in TEX …	all keys	Only replace keys cited in these files (\cite / \citep / \citet / \citealp / \citeyear / \nocite). Unused keys are still pretty-printed and still affected by -r / -s / -st / -d
--live-lookup	off	On a local miss, search DBLP by title (max 5 hits). Same author/title guards. Respect DBLP rate limits
--keep FIELDS	keep all	Allowlist of fields (ID / ENTRYTYPE always kept)
--protect-titles	off	Brace acronyms in titles (BERT, GPT-2, …)
--format-only	off	Pretty-print only; no DB / DBLP / arXiv rewrite
--no-check-authors	off	Skip last-name overlap (empty authors still do not match)
-r	none	Drop fields, e.g. -r url,biburl,timestamp
-s True	False	Shorten venues via rebiber/abbr.tsv
-d False	True	Keep duplicate cite keys
-st True	False	Sort entries by cite key
-l / -a	packaged	Custom bib_list.txt / abbr.tsv
-u	—	Refresh packaged dumps from GitHub main
-v	—	Print version

Leftover unofficial arXiv entries may still query the arXiv API for year / primaryClass. Use --format-only for a fully offline run.

Matching

• Authors. A title hit converts only if the first-author last names match or at least two last names overlap. Empty/missing authors never match (including with --no-check-authors). Trailing et al. / et.al. / and others are stripped.
• Titles. Digit-preserving key first (16x16 ≠ 32x32), then the letters-only key used by older dumps.
• Live DBLP. Prefers a published (non-arXiv/CoRR) hit; skips if several published hits remain. Digit keys must agree.
• eprint. If the input has an arXiv id and the official record does not, it is copied onto the replacement (local dump and live).
• ArXiv form. Unofficial preprints become @misc with eprint / archivePrefix / primaryClass. Already-published papers are not rewritten because an abstract mentions arXiv.
• Broken entries. Unparsed or unclosed records are kept, not dropped.

Build a dump JSON:

uv run python -m rebiber.bib2json -i path/to/conf.bib -o path/to/conf.json

Example

@article{lin2020birds,
  title={Birds have four legs?! NumerSense: Probing Numerical Commonsense Knowledge of Pre-trained Language Models},
  author={Lin, Bill Yuchen and Lee, Seyeon and Khanna, Rahul and Ren, Xiang},
  journal={arXiv preprint arXiv:2005.00683},
  year={2020}
}

becomes the official EMNLP 2020 @inproceedings (same cite key), with anthology URL and DOI.

Supported venues

Toggle files in rebiber/bib_list.txt. ACL Anthology includes *CL main conferences and workshops. Other dumps are main tracks only (no CVPR/ICCV workshops).

A monthly Action opens a PR with fresh DBLP + anthology data. Years below are what is in this repo, not “whatever DBLP has tomorrow.”

Kept current (monthly job + recent dumps)

Venue	Years
ACL Anthology	current (split acl_1/2/3.json)
AAAI	2010–2026
ACCV	2022, 2024
AISTATS	2013–2025
BMVC	2010–2024
CHI	2010–2026
CVPR	2000–2025
ECCV	2022, 2024
ICASSP	2015–2025
ICCV	odd years 2003–2025
ICLR	2013–2025
ICML	2000–2025
IJCAI	2011, 2013, 2015–2025
INTERSPEECH	2016–2025
KDD	2010–2026
ICRA / IROS	2020–2025 (main)
RSS / CoRL	2020–2024 (main)
JMLR	2020–2026
TMLR	2022–2026
WACV	2022–2026
MLSys	2019–2025
MICCAI	2022–2025 (main)
NeurIPS	2000–2025
SIGIR	2010–2026
UAI	2010–2025
WWW	2001–2026

Not in the index: COLM (empty DBLP toc). ICLR/ICML/CVPR 2026, RSS/CoRL 2025, ECCV 2026 — not on DBLP yet.

Historical (frozen ~2020): ALENEX, ASONAM, BigData, CIDR, CIKM, COLT, MM, RecSys, SDM, SIGMOD (through 2022; later years are PACMMOD), SODA, STOC, WSDM.

Thanks to Anton Tsitsulin for the original dump collection.

Adding a venue

uv run python -m rebiber.download_dblp --confs iclr --start-year 2026

That writes rebiber/data/{conf}{year}.bib.json and appends bib_list.txt when needed. PRs welcome.

Star history

In-repo SVGs, refreshed weekly from the GitHub API (no third-party token).

Contact

billyuchenlin@gmail.com · i@yuchenlin.xyz · GitHub issues