Hugging Face's logo

Datasets:
sidneibarbieri
/
topvenues-complete

Tasks:
Text Retrieval
Modalities:
Text
Formats:
parquet
Languages:
English
Size:
100K - 1M
Tags:
cybersecurity
security
bibliography
literature-review
dblp
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
 Community
1
topvenues-complete / README.md
sidneibarbieri's picture
sidneibarbieri
Field-wise deduplication: recover 2,743 BibTeX, 1,483 abstracts, 13 truncated titles
cd72a60 verified
preview code
|
Raw
History Blame Contribute Delete
4.31 kB
metadata 
pretty_name: TopVenues Complete Corpus
license: other
language:
  - en
task_categories:
  - text-retrieval
tags:
  - cybersecurity
  - security
  - bibliography
  - literature-review
  - dblp
size_categories:
  - 100K<n<1M
configs:
  - config_name: default
    data_files:
      - split: train
        path: train-*.parquet

TopVenues Complete Corpus

The widest corpus the TopVenues pipeline produces: 118,817 distinct publications across 40 venues (2017-2026), every record carrying a canonical BibTeX entry and a DBLP key.

Built by a monotonic union of the full-40 and security-20 profiles. Neither was a superset of the other, so the union recovers records and abstracts that either profile alone was missing.

Not a paper denominator

This corpus is a tool release, not the denominator of any paper. Two separate frozen snapshots back the published work and are not replaced by this one:

Release Records Cited by
submitted-11 9,925 the full paper
security-20 (dataset sidneibarbieri/topvenues) 20,305 the tools-track paper

Any measurement must name the snapshot that produced its denominator. Numbers from one corpus do not transfer to another merely because the software is shared.

One record per work

Records are deduplicated by DBLP key, field-wise: where the same key was ingested more than once, each field keeps its most complete value, so the surviving record is at least as complete as any copy that fed it. That recovered 2,743 BibTeX entries, 1,483 abstracts, 415 author lists and 13 titles truncated at superscripts. Result: 118,817 rows, 118,817 distinct keys, so a count of rows is a count of distinct publications.

Abstract coverage is uneven, by construction

14,289 of 118,817 records carry an abstract (12.0%). That figure is low because the corpus deliberately includes large AI and systems venues whose records were ingested for bibliographic completeness without abstract enrichment. Coverage on the security venues is high. 21 of 40 venues carry abstracts at all:

Venue Records Abstracts Coverage
USENIX Security 2,103 2,056 97.8%
ACM CCS 1,972 1,971 99.9%
TrustCom 1,915 1,912 99.8%
ACM Computing Surveys 1,813 1,812 99.9%
IEEE S&P 1,166 1,165 99.9%
NDSS 1,059 1,058 99.9%
ACM ASIA CCS 837 827 98.8%
IEEE Communications Surveys & Tutorials 833 804 96.5%
ACSAC 509 509 100.0%
IEEE CNS 381 381 100.0%
IEEE EURO S&P 339 339 100.0%
ACM CODASPY 316 281 88.9%
ACM WiSec 314 270 86.0%
HotNets 245 245 100.0%
RAID 290 222 76.6%
ACM SACMAT 176 176 100.0%
IEEE SaTML 127 127 100.0%
ACM AISec 77 73 94.8%
ESORICS 598 39 6.5%
Foundations and Trends in Privacy and Security 20 20 100.0%
USENIX WOOT 18 2 11.1%
AAAI 20,085 0 0.0%
NeurIPS 16,527 0 0.0%
EMNLP 12,817 0 0.0%
ICML 11,971 0 0.0%
ACL 11,294 0 0.0%
ICLR 10,680 0 0.0%
IJCAI 6,501 0 0.0%
ACM KDD 4,110 0 0.0%
NAACL 4,107 0 0.0%
ACM MobiCom 993 0 0.0%
ACM SenSys 808 0 0.0%
ACM MobiSys 655 0 0.0%
USENIX NSDI 542 0 0.0%
ACM EuroSys 518 0 0.0%
USENIX ATC 518 0 0.0%
ACM IMC 466 0 0.0%
ACM SIGCOMM 452 0 0.0%
ACM SIGMETRICS 376 0 0.0%
ACM HotMobile 289 0 0.0%

Usage 

from datasets import load_dataset

corpus = load_dataset("sidneibarbieri/topvenues-complete", split="train")
security = corpus.filter(lambda r: r["abstract"] and "USENIX" in (r["event"] or ""))

Fields

paper_id, key (DBLP), title, authors, event, venue, area, year, paper_type, pages, url, ee (DOI), abstract, bibtex.

License

Metadata originates from DBLP (CC0) and open scholarly APIs. Abstracts remain under their publishers' terms and are included for research use.