ERASE: A Real-World Aligned Benchmark for Unlearning in Recommender Systems

ERASE is a benchmark for evaluating machine unlearning methods in recommender systems. It provides a standardized framework for measuring unlearning efficiency, utility, and effectiveness across multiple datasets and models. The dataset was created by deem-data and last updated on 2026-02 18.

Use Cases

• Benchmarking unlearning efficiency based on the standardized framework
• Evaluating utility preservation in recommender systems after unlearning
• Comparing effectiveness of different unlearning methods across multiple datasets
• Testing real-world inspired conditions for unlearning algorithms

Strengths

• Provides a standardized framework for evaluation across multiple datasets and models
• Focuses on real-world inspired conditions for machine unlearning
• Includes trained, retrained, and unlearned model states as indicated by repository contents

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

deem-data on Hugging Face

Collection Method

Likely contains trained, retrained, and unlearned model states for benchmarking.

Time Range

null

Freshness

Last updated 2026-02-18 16:46:55

Geography

null