| { |
| "@context": { |
| "@language": "en", |
| "@vocab": "https://schema.org/", |
| "arrayShape": "cr:arrayShape", |
| "citeAs": "cr:citeAs", |
| "column": "cr:column", |
| "conformsTo": "dct:conformsTo", |
| "containedIn": "cr:containedIn", |
| "cr": "http://mlcommons.org/croissant/", |
| "data": { |
| "@id": "cr:data", |
| "@type": "@json" |
| }, |
| "dataBiases": "cr:dataBiases", |
| "dataCollection": "cr:dataCollection", |
| "dataType": { |
| "@id": "cr:dataType", |
| "@type": "@vocab" |
| }, |
| "dct": "http://purl.org/dc/terms/", |
| "extract": "cr:extract", |
| "field": "cr:field", |
| "fileProperty": "cr:fileProperty", |
| "fileObject": "cr:fileObject", |
| "fileSet": "cr:fileSet", |
| "format": "cr:format", |
| "includes": "cr:includes", |
| "isArray": "cr:isArray", |
| "isLiveDataset": "cr:isLiveDataset", |
| "jsonPath": "cr:jsonPath", |
| "key": "cr:key", |
| "md5": "cr:md5", |
| "parentField": "cr:parentField", |
| "path": "cr:path", |
| "personalSensitiveInformation": "cr:personalSensitiveInformation", |
| "recordSet": "cr:recordSet", |
| "references": "cr:references", |
| "regex": "cr:regex", |
| "repeated": "cr:repeated", |
| "replace": "cr:replace", |
| "sc": "https://schema.org/", |
| "separator": "cr:separator", |
| "source": "cr:source", |
| "subField": "cr:subField", |
| "transform": "cr:transform", |
| "rai": "http://mlcommons.org/croissant/RAI/", |
| "prov": "http://www.w3.org/ns/prov#" |
| }, |
| "@type": "sc:Dataset", |
| "distribution": [ |
| { |
| "@type": "cr:FileObject", |
| "@id": "repo", |
| "name": "repo", |
| "description": "The Hugging Face git repository.", |
| "contentUrl": "https://huggingface.co/datasets/logo-lab/trl-rbench/tree/refs%2Fconvert%2Fparquet", |
| "encodingFormat": "git+https", |
| "sha256": "https://github.com/mlcommons/croissant/issues/80" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "parquet-files-for-config-record_linkage", |
| "containedIn": { |
| "@id": "repo" |
| }, |
| "encodingFormat": "application/x-parquet", |
| "includes": "record_linkage/*/*.parquet" |
| }, |
| { |
| "@type": "cr:FileSet", |
| "@id": "parquet-files-for-config-row_prediction", |
| "containedIn": { |
| "@id": "repo" |
| }, |
| "encodingFormat": "application/x-parquet", |
| "includes": "row_prediction/*/*.parquet" |
| } |
| ], |
| "recordSet": [ |
| { |
| "@type": "cr:RecordSet", |
| "dataType": "cr:Split", |
| "key": { |
| "@id": "record_linkage_splits/split_name" |
| }, |
| "@id": "record_linkage_splits", |
| "name": "record_linkage_splits", |
| "description": "Splits for the record_linkage config.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage_splits/split_name", |
| "dataType": "sc:Text" |
| } |
| ], |
| "data": [ |
| { |
| "record_linkage_splits/split_name": "train" |
| }, |
| { |
| "record_linkage_splits/split_name": "validation" |
| }, |
| { |
| "record_linkage_splits/split_name": "test" |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "record_linkage", |
| "description": "logo-lab/trl-rbench - 'record_linkage' subset\n\nAdditional information:\n- 3 splits: train, validation, test", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/split", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "fileProperty": "fullpath" |
| }, |
| "transform": { |
| "regex": "record_linkage/(?:partial-)?(train|validation|test)/.+parquet$" |
| } |
| }, |
| "references": { |
| "field": { |
| "@id": "record_linkage_splits/split_name" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/source", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "column": "source" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/family", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "column": "family" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/pair_id", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "column": "pair_id" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/table_a_record_json", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "column": "table_a_record_json" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/table_b_record_json", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "column": "table_b_record_json" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "record_linkage/label", |
| "dataType": "cr:Int32", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-record_linkage" |
| }, |
| "extract": { |
| "column": "label" |
| } |
| } |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "dataType": "cr:Split", |
| "key": { |
| "@id": "row_prediction_splits/split_name" |
| }, |
| "@id": "row_prediction_splits", |
| "name": "row_prediction_splits", |
| "description": "Splits for the row_prediction config.", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction_splits/split_name", |
| "dataType": "sc:Text" |
| } |
| ], |
| "data": [ |
| { |
| "row_prediction_splits/split_name": "train" |
| }, |
| { |
| "row_prediction_splits/split_name": "validation" |
| }, |
| { |
| "row_prediction_splits/split_name": "test" |
| } |
| ] |
| }, |
| { |
| "@type": "cr:RecordSet", |
| "@id": "row_prediction", |
| "description": "logo-lab/trl-rbench - 'row_prediction' subset\n\nAdditional information:\n- 3 splits: train, validation, test", |
| "field": [ |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/split", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "fileProperty": "fullpath" |
| }, |
| "transform": { |
| "regex": "row_prediction/(?:partial-)?(train|validation|test)/.+parquet$" |
| } |
| }, |
| "references": { |
| "field": { |
| "@id": "row_prediction_splits/split_name" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/openml_id", |
| "dataType": "cr:Int32", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "openml_id" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/dataset_name", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "dataset_name" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/row_idx", |
| "dataType": "cr:Int32", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "row_idx" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/record_json", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "record_json" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/targets_json", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "targets_json" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/target_specs_json", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "target_specs_json" |
| } |
| } |
| }, |
| { |
| "@type": "cr:Field", |
| "@id": "row_prediction/dataset_metadata_json", |
| "dataType": "sc:Text", |
| "source": { |
| "fileSet": { |
| "@id": "parquet-files-for-config-row_prediction" |
| }, |
| "extract": { |
| "column": "dataset_metadata_json" |
| } |
| } |
| } |
| ] |
| } |
| ], |
| "conformsTo": "http://mlcommons.org/croissant/1.1", |
| "name": "trl-rbench", |
| "description": "\n\t\n\t\t\n\t\tTRL-Rbench\n\t\n\nRow-level evaluation suite of TRL-Bench:\n\nrow_prediction: 50 OpenML tables, 1.1M rows (887,720 train / 110,962 validation / 110,983 test) across 123 hand-verified targets. Per-row schema bundles features and targets as JSON dicts so all 50 tables share one config.\nrecord_linkage: 16 entity-matching sources (8 clean DeepMatcher + 4 dirty DeepMatcher + 4 WDC LSPM v2 sizes) unified into one config with 357,833 train / 95,817 validation / 42,835 test row pairs. Per-row stores… See the full description on the dataset page: https://huggingface.co/datasets/logo-lab/trl-rbench.", |
| "alternateName": [ |
| "logo-lab/trl-rbench", |
| "TRL-Rbench" |
| ], |
| "creator": { |
| "@type": "Organization", |
| "name": "LOGO Lab", |
| "url": "https://huggingface.co/logo-lab" |
| }, |
| "keywords": [ |
| "English", |
| "cc-by-4.0", |
| "1M - 10M", |
| "parquet", |
| "Tabular", |
| "Text", |
| "Datasets", |
| "Dask", |
| "Polars", |
| "Croissant", |
| "🇺🇸 Region: US", |
| "tabular", |
| "row-level", |
| "benchmark", |
| "representation-learning", |
| "entity-matching", |
| "record-linkage", |
| "trl-bench" |
| ], |
| "license": "https://choosealicense.com/licenses/cc-by-4.0/", |
| "url": "https://huggingface.co/datasets/logo-lab/trl-rbench", |
| "rai:dataLimitations": "TRL-Bench is designed to standardize cross-paradigm representation-level comparison of tabular encoders under shared lightweight downstream readouts, not to report best end-to-end systems. Scores therefore answer a narrower question — what common lightweight readouts can extract from exported embeddings — rather than replacing fully adapted-system benchmarks. The protocol standardizes task definitions and downstream evaluation, not raw model preprocessing: each encoder runs in its documented operating regime through its standard supported wrapper rather than a single forced serialization, which improves fidelity to the original models but leaves some wrapper-induced variation. Reproducing all 20 models across the suites is computationally nontrivial. The current release does not cover temporal, multimodal, or highly specialized scientific tables. Several assets are design choices rather than natural distributions: row-prediction targets are manually curated from OpenML candidates (50 of 158 retained), and TRL-DLTE is a synthetic fragmentation benchmark derived from TabFact and WikiTableQuestions parent tables rather than a naturally-occurring enrichment workload. Dataset counts and normalized-rank summaries are convenience aggregates and should be read alongside the per-task results.", |
| "rai:dataBiases": "Source distribution skews toward English-language Web and Wikipedia content (SATO/VizNet, SOTAB/Web Data Commons, WikiCT/TURL, TabFact, WikiTableQuestions, NQ-Tables, WDC LSPM, LakeBench Wiki/CKAN/Spider). Manual selection bias affects row prediction (50 of 158 candidate OpenML tables retained after filtering for non-degenerate multi-target structure). DLTE distractor pool draws from CKAN open-government portals with geographic skew toward Western jurisdictions (US, Canada, UK, Singapore). Models in the empirical study lean on existing pretraining corpora that themselves have known representational biases. The benchmark does not include explicit demographic-fairness probes; auditing fairness on downstream tasks built on these representations remains open work.", |
| "rai:personalSensitiveInformation": "No personally identifying information is directly redistributed. Upstream content sources include public Wikipedia tables (people, places, events; license-compliant under CC-BY-SA), public Web Data Commons schema.org tables, public OpenML benchmark tables (commonly CC0/CC-BY), and open-government CKAN portals. WDC Products tables contain product titles, categories, and prices from e-commerce sites. Label-equivalent columns that could leak entity-matching identity (`cluster_id`, `identifiers` on WDC; `class` on Fodors-Zagats) have already been removed at the parquet layer before any encoder consumes a row. The benchmark itself is not designed for re-identification; downstream record-linkage applications must consult the original dataset terms before deploying in privacy-sensitive contexts.", |
| "rai:dataUseCases": "Validated use case: evaluation of frozen tabular encoder representations under a shared cross-paradigm protocol; comparing row-, column-, or table-level embeddings exported by heterogeneous models on standardized lightweight downstream readouts. Not validated for production deployment of any individual encoder, for use as training data in foundation-model pretraining, or for large-scale data linkage in privacy-sensitive settings — the paper explicitly raises surveillance, re-identification, and inappropriate dataset fusion as concerns for downstream record-linkage and data-lake retrieval applications. The `record_linkage` config supports row-pair entity-matching evaluation; downstream production deployment in privacy-sensitive linkage scenarios is explicitly out of scope.", |
| "rai:dataSocialImpact": "Positive impact: improves scientific comparability across tabular encoder paradigms, reduces evaluation fragmentation, and helps practitioners identify when specialized table-aware models are necessary vs. when simpler frozen encoders suffice. Negative impact: stronger table representations can be misused for surveillance, re-identification, and inappropriate dataset fusion, particularly in record-linkage and data-lake retrieval settings. Mitigations applied: per-config license enforcement and inheritance documentation; pre-removal of label-equivalent leak columns; parent-disjoint and pair-disjoint train/dev/test splits; explicit RAI metadata in this Croissant file; an annotated license map (LICENSES.md) per repository documenting upstream redistribution constraints.", |
| "rai:hasSyntheticData": false, |
| "prov:wasDerivedFrom": [ |
| { |
| "@id": "https://www.openml.org/" |
| }, |
| { |
| "@id": "https://github.com/autogluon/tabarena" |
| }, |
| { |
| "@id": "https://www.openml.org/s/99" |
| }, |
| { |
| "@id": "https://www.openml.org/s/353" |
| }, |
| { |
| "@id": "https://pages.cs.wisc.edu/~anhai/data1/deepmatcher_data/" |
| }, |
| { |
| "@id": "https://webdatacommons.org/largescaleproductcorpus/v2/" |
| } |
| ], |
| "prov:wasGeneratedBy": "TRL-Bench (NeurIPS 2026 E&D submission) curation pipeline: source datasets are fetched from the upstream URLs listed in `prov:wasDerivedFrom`, normalized to a shared per-config schema, and re-released as parquet on HuggingFace. Curation steps include: standardizing table identifiers and label schemas; rewriting end-to-end source tasks into representation-centric variants (e.g., DeepMatcher row-pair rewrites); enforcing table-disjoint splits for joint pair tasks; manual review and label repair on the OpenML row-prediction targets (50 of 158 candidates retained, 123 hand-verified targets); fragmenting TabFact + WTQ parent tables into the synthetic 47,772-table DLTE retrieval lake at four cumulative noise tiers (clean / schema / cell / hard); pre-removing label-equivalent leak columns (`cluster_id`, `identifiers` on WDC; `class` on Fodors-Zagats) before any encoder consumes a row.", |
| "citeAs": "@inproceedings{trl_bench_2026,\n title = {TRL-Bench: Standardizing Cross-Paradigm Representation-Level Evaluation of Tabular Encoders},\n author = {Anonymous Authors},\n year = {2026},\n booktitle = {NeurIPS Evaluations and Datasets Track},\n}", |
| "datePublished": "2026-05-04", |
| "version": "1.0.0" |
| } |
