Hugging Face's logo

Datasets:
graphguard-anon
/
graphguard-benchmarks

Modalities:
Text
Tags:
database-integrity
automated-reasoning
first-order-logic
craig-interpolation
formal-verification
sigmod-reproducibility
License:
Dataset card Data Studio Files
xet
 Community
graphguard-benchmarks / README.md
graphguard-anon's picture
graphguard-anon
Add anonymized benchmark datasets and constraint suites for paper evaluation
fc7120d verified
preview code
|
Raw History Blame Contribute Delete
3.73 kB
metadata 
license: apache-2.0
task_categories:
  - graph-machine-learning
  - relational-databases
  - knowledge-graphs
tags:
  - database-integrity
  - automated-reasoning
  - first-order-logic
  - craig-interpolation
  - formal-verification
  - sigmod-reproducibility
size_categories:
  - 10GB<n<100GB

UpdateGuard Benchmark Suite (Anonymized Submission)

This repository contains the complete cross-paradigm benchmark suites, schemas, constraints, and data instances for UpdateGuard, an automated first-order theorem proving and Craig interpolation framework for database integrity verification.

Repository Contents

The repository is modularly structured into tier-based packages:

1. Constraints Suite (constraints.zip ~0.5 MB)

Contains all integrity constraints evaluated across the three paradigms:

  • Property Graphs: Mined and hand-crafted Graph Generating Dependencies (GGDs) for CORDIS, DBLP, GDELT, and LDBC SNB, alongside property-level constraints: Graph Object Functional Dependencies (GO-FDs) for Northwind, graph functional/uniqueness dependencies (gFDs/gUCs) for Offshore Leaks, and graph functional dependencies and keys for London Public Transport (LPT) and Train Services (TS).
  • RDF Knowledge Graphs: W3C SHACL shape schemas and OWL ontologies for LUBM and WatDiv.
  • Relational Schemas: Relational dependencies (foreign keys, uniqueness EGDs, check constraints) for TPC-H and all ChaseBench scenarios (doctors, doctors-fd, LUBM, Ontology-256, STB-128, deep, and correctness suites).

2. Standard Evaluation Benchmarks (graph_benchmarks.zip, rdf_benchmarks.zip, relational_benchmarks.zip)

  • Property Graphs (graph_benchmarks.zip): Ready-to-evaluate node and edge CSV files for Neo4j and localized r-hop neighborhood extraction (cordis_small, dblp_small, gdelt_small, ldbc_small_*, northwind, offshore_small, lpt, trains).
  • RDF Knowledge Graphs (rdf_benchmarks.zip): LUBM multi-university instances (1 to 15 universities in OWL/RDF), WatDiv 100k triple sample (watdiv_100k.nt), and UBA synthetic data generator.
  • Relational Databases (relational_benchmarks.zip): ChaseBench relational scenarios with tables, schemas, queries, and repair tools, alongside TPC-H decision-support tables at scale factors 0.01 to 0.1.

3. Full-Scale Datasets (Optional, for heavy data scaling)

  • ldbc_scales.zip: LDBC SNB SF 1 and SF 3 instances.
  • watdiv_10m.zip: WatDiv full 10M triples dataset (watdiv.10M.nt).
  • dblp_raw.zip: Complete DBLP publication XML database (dblp.xml, dblp.dtd).
  • tpch_scales.zip: TPC-H tables at SF 0.3, SF 1, SF 3, and SF 10.

Quick Download & Setup

You can download and extract these datasets directly into the codebase using the automated downloader script:

# Download standard evaluation bundle (constraints + benchmark datasets):
python scripts/download_benchmarks.py

# Download only constraints (ultra-lightweight, ~300 KB):
python scripts/download_benchmarks.py --constraints-only

# Download specific datasets:
python scripts/download_benchmarks.py --dataset ldbc cordis

All archives extract into the datasets/ root directory, maintaining the exact folder paths expected by the evaluation pipelines (run_all_benchmarks.py, run_eval.sh, run_chasebench_eval.sh, and run_lubm_scaling.sh).

Double-Blind Anonymity Compliance

This repository and its contents are fully anonymized for double-blind peer review. All local paths, personal metadata, and author identifications have been scrubbed in accordance with SIGMOD review guidelines.