Auto57%Harbor: pendingKnowledge & Reasoning

FASHION-RQBENCH

FASHION-RQ: Mining Hierarchical Semantic Structure for Automated Evaluation of Fashion Retrieval

A scalable fashion retrieval benchmark with 5,380 queries and 44,446 products, evaluating ranking, relevance, sibling discrimination, and paraphrase robustness using hierarchical semantic structures.

How to run

ltb · fashion-rqbench
$ npm i -g lettokenburn-cli
$ ltb bench info fashion-rqbench

External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.

Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs

Scores

9 sourced entries
HarnessSourceConfidenceLink
Qwen3-Embedding-4B0.9N/AIndependentmediumOct 13, 2026arxiv.org
E5-Mistral-7B-Instruct0.9N/AIndependentmediumOct 13, 2026arxiv.org
Qwen3-Embedding-8B0.9N/AIndependentmediumOct 13, 2026arxiv.org
SimCSE-RoBERTa-large0.9N/AIndependentmediumOct 13, 2026arxiv.org
Contriever-MSMARCO0.9N/AIndependentmediumOct 13, 2026arxiv.org
SPLADE++0.9N/AIndependentmediumOct 13, 2026arxiv.org
ColBERTv20.9N/AIndependentmediumOct 13, 2026arxiv.org
Sentence-T5-large0.8N/AIndependentmediumOct 13, 2026arxiv.org
RoBERTa-large0.7N/AIndependentmediumOct 13, 2026arxiv.org

Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.

Reproduced by the community

All runs
Loading community runs…

About

FASHION-RQBENCH comprises 5,380 attribute-enriched queries and 44,446 products derived from catalog structure via RQ-VAE. It supports four tasks: ranking, hierarchical relevance classification, sibling discrimination, and paraphrase robustness.

Scoring uses weighted query-relevance and per-attribute metrics. Experiments show dense encoders outperform BM25 in ranking (nDCG@10 0.87 vs 0.77) but struggle with fine-grained attributes like material and style.