BenchPreS: A Benchmark for Personalized Preference Selectivity in LLMs

BenchPreS is a benchmark for evaluating persistent-memory large language models. It pairs 10 user profiles with 39 recipient-task contexts across five formal communication domains. The dataset was created by sangyon and last updated on March 20, 2026.

Use Cases

• Benchmarking LLM personalization capabilities based on stored user profiles
• Evaluating context-aware preference suppression based on recipient-task contexts
• Testing LLM performance across five formal communication domains

Strengths

• Includes 10 distinct user profiles containing factual and preference attributes
• Tests across 39 recipient-task contexts
• Covers five formal communication domains

Limitations

• Column-level documentation is absent; field semantics must be inferred after download
• Row count is unknown, which may limit suitability assessment

Provenance

Source

huggingface

Freshness

Last updated 2026-03-20 02:27:52