OBLIQ-Bench: A Suite of Five Retrieval Benchmarks for Latent Queries

OBLIQ-Bench is a suite of five retrieval benchmarks designed to expose a blind spot in modern search systems. It focuses on oblique queries where relevance-determining attributes are latent and have little surface expression in documents. The dataset was created by author 'dianetc' and last updated on Hugging Face in May 2026.

Use Cases

• Evaluating retrieval system performance on queries with latent relevance criteria.
• Benchmarking the ability of search systems to find documents based on implied attributes.
• Training or fine-tuning retrieval models to handle oblique query scenarios.
• Researching the gap between document recognition and retrieval from large corpora.

Strengths

• Specifically designed to test a known blind spot in modern retrieval systems.
• Comprises five distinct benchmarks, suggesting a multi-faceted evaluation approach.
• Relevance can be verified by a reasoning LLM, providing a potential ground-truth method.

Limitations

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

Provenance

Source

Hugging Face dataset by author dianetc.

Collection Method

Likely constructed for research purposes to benchmark retrieval systems.

Time Range

null

Freshness

Last updated 2026-05-06 15:09:36; freshness should be verified.

Geography

null