CrossER: A Benchmark for Context-Dependent Cross-System Entity Resolution

CrossER is a benchmark dataset for context-dependent cross-system entity resolution where surface features are deliberately misleading. Match pairs average only 0.29 string similarity, while non-match pairs average 0.94 similarity, simulating real enterprise scenarios. The dataset was created by author smurthy5 and was last updated on Hugging Face in June 2026.

Use Cases

• Benchmarking entity resolution algorithms based on the described low string similarity for true matches.
• Training models for cross-system product matching based on the context-dependent nature of the data.
• Studying the limitations of surface-level feature matching based on the described high similarity for non-matches.
• Simulating enterprise data migration scenarios based on the description referencing migration runbooks and classification guides.

Strengths

• Designed as a benchmark with specific, measurable characteristics: match pairs average 0.29 string similarity.
• Simulates a realistic, challenging scenario: non-match pairs average 0.94 string similarity, making them appear identical.
• Focuses on context-dependent resolution, a key aspect of real-world enterprise data integration.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

Hugging Face user smurthy5.

Collection Method

Likely constructed as a benchmark for entity resolution research.

Freshness

Last updated 2026-06-22 02:23:50; freshness should be verified.