Upload croissant.json with huggingface_hub

croissant.json

@@ -0,0 +1,367 @@
+{
+  "@context": {
+    "@language": "en",
+    "@vocab": "https://schema.org/",
+    "citeAs": "cr:citeAs",
+    "column": "cr:column",
+    "conformsTo": "dct:conformsTo",
+    "cr": "http://mlcommons.org/croissant/",
+    "data": {"@id": "cr:data", "@type": "@json"},
+    "dataType": {"@id": "cr:dataType", "@type": "@vocab"},
+    "dct": "http://purl.org/dc/terms/",
+    "extract": "cr:extract",
+    "field": "cr:field",
+    "fileProperty": "cr:fileProperty",
+    "fileSet": "cr:fileSet",
+    "format": "cr:format",
+    "includes": "cr:includes",
+    "isLiveDataset": "cr:isLiveDataset",
+    "jsonPath": "cr:jsonPath",
+    "key": "cr:key",
+    "md5": "cr:md5",
+    "parentField": "cr:parentField",
+    "path": "cr:path",
+    "rai": "http://mlcommons.org/croissant/RAI/",
+    "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"
+  },
+  "@type": "sc:Dataset",
+  "name": "CrossER",
+  "description": "CrossER is a benchmark for context-dependent cross-system entity resolution. It measures how well systems can leverage messy enterprise documents (migration trackers, classification guides, Slack threads) to resolve entities across 5 disconnected enterprise systems with different naming conventions (English, German, abbreviated codes, legacy codes). Match pairs average only 0.29 string similarity while non-match pairs average 0.94, making surface features deliberately misleading. The benchmark provides three evaluation modes: no context, raw enterprise documents (8 signal + 110 noise), and oracle structured context.",
+  "conformsTo": "http://mlcommons.org/croissant/1.0",
+  "license": "https://creativecommons.org/licenses/by/4.0/",
+  "url": "https://huggingface.co/datasets/smurthy5/CrossER",
+  "version": "2.0.0",
+  "citeAs": "@inproceedings{crosser2026, author={Gunukula, Nihal and Murthy, Sameer}, title={{CrossER: A Benchmark for Context-Dependent Cross-System Entity Resolution}}, booktitle={NeurIPS 2026 Evaluations \\& Datasets Track}, year={2026}, url={https://huggingface.co/datasets/smurthy5/CrossER}}",
+  "datePublished": "2026-03-23",
+  "creator": [
+    {
+      "@type": "Person",
+      "name": "Nihal Gunukula",
+      "url": "https://github.com/nihalgunu"
+    },
+    {
+      "@type": "Person",
+      "name": "Sameer Murthy",
+      "url": "https://github.com/SameerMurthy5"
+    }
+  ],
+  "publisher": {
+    "@type": "Organization",
+    "name": "Phyvant",
+    "url": "https://phyvant.com"
+  },
+  "keywords": [
+    "entity resolution",
+    "record linkage",
+    "cross-system matching",
+    "enterprise data",
+    "context-dependent",
+    "multilingual",
+    "benchmark",
+    "NLP",
+    "RAG",
+    "information extraction"
+  ],
+
+  "rai:dataCollection": "Synthetically generated using large language models (GPT-4 class) to simulate realistic enterprise entity resolution scenarios across five enterprise domains: commodity trading, insurance underwriting, logistics operations, financial compliance, and vendor management. Generation is fully reproducible via seed=42.",
+  "rai:dataCollectionType": "Synthetic",
+  "rai:dataCollectionRawData": "Raw data consists of synthetically generated enterprise documents including migration trackers, classification guides, Slack-style threads, email chains, support tickets, ERP event logs, and knowledge documents. These simulate real enterprise data without containing any actual enterprise records.",
+  "rai:dataCollectionTimeFrame": "Generated in 2025–2026 for the NeurIPS 2026 Evaluations & Datasets Track submission.",
+  "rai:dataCollectionMissingData": "No missing data. All entity pairs have ground truth labels. Context documents are intentionally incomplete to simulate realistic RAG scenarios where relevant information may be absent.",
+  "rai:dataPreprocessingProtocol": "Data is generated deterministically using seed=42 via the included generation pipeline (generate/generate_all.py). No manual preprocessing is applied beyond the generation pipeline. Context documents are split into signal (8 documents containing entity mapping information) and noise (110 documents irrelevant to entity resolution) to simulate a realistic enterprise RAG corpus.",
+  "rai:dataPreprocessingManipulation": "Entity names are deliberately varied across systems to reflect real enterprise naming conventions: formal English (SAP_TC2), internal codes/abbreviations (SAP_CFIN), region-prefixed abbreviations (SAP_APAC), cryptic legacy codes (LEGACY_ERP), and authoritative long-form names (SHAREPOINT). German-language names are included in select entity types.",
+  "rai:dataAnnotationProtocol": "Ground truth labels (match / no_match / ambiguous) and difficulty tiers (easy, medium, hard, adversarial_negative, medium_negative, obvious_negative, ambiguous) were assigned programmatically by the dataset authors using a deterministic rule-based system with seed=42. Labels are derived from the canonical entity mapping table (_canonical_mappings.json) that defines the true cross-system entity equivalences.",
+  "rai:dataAnnotationPlatform": "Programmatic annotation by the dataset authors; no crowdsourcing platform was used.",
+  "rai:dataAnnotationAnalysis": "All annotations are derived deterministically from the canonical mapping table. Pair difficulty tiers are assigned based on string similarity thresholds and entity type combinations to create a controlled difficulty distribution: easy (high-similarity matches + obvious negatives), medium (moderate-difficulty pairs), hard (low-similarity matches + adversarial high-similarity negatives + ambiguous cases).",
+  "rai:annotationsPerItem": "1",
+  "rai:annotatorDemographics": "Labels are assigned programmatically by the dataset authors; no human crowd-sourced annotators were used. The authors are graduate researchers in database systems and NLP.",
+  "rai:machineAnnotationTools": "Data generation pipeline using GPT-4 class models (OpenAI API) with deterministic seed=42. Ground truth labels are generated programmatically from the canonical mapping table without LLM involvement to avoid label noise.",
+  "rai:personalSensitiveInformation": "None. The dataset is entirely synthetic and contains no real personal, financial, or sensitive enterprise information. All entity names, company references, and document content are fictitious.",
+  "rai:dataBiasAnalysis": "CrossER intentionally introduces surface-feature bias to stress-test entity resolution systems: match pairs have mean string similarity 0.29 (names look unrelated) while non-match pairs have mean string similarity 0.94 (names look identical). This design bias is intentional and documented. Domain distribution is balanced across 5 enterprise domains. Entity type distribution is balanced across 4 types (product, supplier, tax_code, legal_entity). Language distribution includes English (primary) and German (secondary, in select entity types).",
+  "rai:dataLimitations": "CrossER is synthetically generated and may not capture all nuances of real-world enterprise data quality issues. The benchmark covers five specific enterprise domains and may not generalize to all industry verticals. Context document quality simulates but does not replicate actual enterprise document heterogeneity. All entities are English or German; other languages are not represented. The benchmark does not cover temporal entity resolution (entity changes over time).",
+  "rai:dataSocialImpact": "CrossER advances research in enterprise entity resolution, which improves data quality, reduces duplicate records, and enhances automated decision-making in organizations. Improved entity resolution systems can reduce manual data reconciliation labor and improve downstream analytics accuracy. The benchmark specifically addresses the underexplored problem of context-dependent matching, which is critical for enterprise AI systems.",
+  "rai:dataUseCases": "Benchmarking entity resolution and record linkage systems; evaluating context-augmented matching (RAG-based ER); studying cross-lingual and cross-system entity matching; evaluating large language model performance on structured matching tasks; developing and evaluating information extraction systems for enterprise knowledge graphs; training and evaluating retrieval systems for enterprise RAG pipelines.",
+
+  "distribution": [
+    {
+      "@type": "cr:FileObject",
+      "@id": "entities-file",
+      "name": "entities.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/entities.json",
+      "encodingFormat": "application/json",
+      "description": "688 entity records from 5 enterprise systems (SAP_TC2, SAP_CFIN, SAP_APAC, LEGACY_ERP, SHAREPOINT) across 4 entity types"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "pairs-file",
+      "name": "pairs.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/pairs.json",
+      "encodingFormat": "application/json",
+      "description": "1,800 entity pair judgments with ground truth labels and 7-tier difficulty annotations"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "train-split",
+      "name": "splits/train.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/splits/train.json",
+      "encodingFormat": "application/json",
+      "description": "Training split: 719 pairs (40% of total)"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "val-split",
+      "name": "splits/val.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/splits/val.json",
+      "encodingFormat": "application/json",
+      "description": "Validation split: 359 pairs (20% of total)"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "test-split",
+      "name": "splits/test.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/splits/test.json",
+      "encodingFormat": "application/json",
+      "description": "Test split: 722 pairs (40% of total)"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "subset-easy",
+      "name": "subsets/crosser_easy.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/subsets/crosser_easy.json",
+      "encodingFormat": "application/json",
+      "description": "CrossER-Easy: 257 pairs (easy matches + obvious negatives); attribute-matching baseline ceiling F1=1.000"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "subset-medium",
+      "name": "subsets/crosser_medium.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/subsets/crosser_medium.json",
+      "encodingFormat": "application/json",
+      "description": "CrossER-Medium: 262 medium-difficulty pairs; attribute-matching baseline ceiling F1=0.776"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "subset-hard",
+      "name": "subsets/crosser_hard.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/subsets/crosser_hard.json",
+      "encodingFormat": "application/json",
+      "description": "CrossER-Hard: 203 pairs (hard matches + adversarial negatives + ambiguous); no-context baseline F1=0.000"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "subset-full",
+      "name": "subsets/crosser_full.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/subsets/crosser_full.json",
+      "encodingFormat": "application/json",
+      "description": "CrossER-Full: 722 test pairs (all difficulty tiers combined)"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "oracle-context-file",
+      "name": "context/structured/oracle_context.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/context/structured/oracle_context.json",
+      "encodingFormat": "application/json",
+      "description": "875 structured oracle context records with complete canonical entity mappings across systems"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "alias-table",
+      "name": "context/structured/alias_table.json",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/context/structured/alias_table.json",
+      "encodingFormat": "application/json",
+      "description": "Cross-system alias lookup table mapping entity names across all 5 source systems"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "raw-signal-documents",
+      "name": "raw_signal_documents",
+      "containedIn": {"@id": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/context/raw/"},
+      "includes": "documents/*.{txt,md,pdf}",
+      "encodingFormat": "text/plain",
+      "description": "8 signal documents containing entity mapping information (migration trackers, classification guides, Slack threads)"
+    },
+    {
+      "@type": "cr:FileSet",
+      "@id": "raw-noise-documents",
+      "name": "raw_noise_documents",
+      "containedIn": {"@id": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/context/raw/"},
+      "includes": "noise/*.{txt,md}",
+      "encodingFormat": "text/plain",
+      "description": "110 noise documents (IT tickets, HR policies, expense reports, meeting notes) irrelevant to entity resolution"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "email-chains",
+      "name": "data/email_chains/",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/email_chains/",
+      "encodingFormat": "application/json",
+      "description": "50 synthetic enterprise email threads with entity references and ground truth"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "support-tickets",
+      "name": "data/tickets/",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/tickets/",
+      "encodingFormat": "application/json",
+      "description": "100 support tickets with resolution chains and entity mentions"
+    },
+    {
+      "@type": "cr:FileObject",
+      "@id": "erp-events",
+      "name": "data/erp_events/",
+      "contentUrl": "https://huggingface.co/datasets/smurthy5/CrossER/resolve/main/data/erp_events/",
+      "encodingFormat": "application/json",
+      "description": "500 cross-system ERP transaction event logs across 15 source systems"
+    }
+  ],
+
+  "recordSet": [
+    {
+      "@type": "cr:RecordSet",
+      "@id": "entities",
+      "name": "entities",
+      "description": "Entity records from 5 enterprise systems with system-specific naming conventions and attributes",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "entities/entity_id",
+          "name": "entity_id",
+          "dataType": "sc:Text",
+          "description": "Unique entity identifier (e.g., SAP_TC2_PROD_001)"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entities/source_system",
+          "name": "source_system",
+          "dataType": "sc:Text",
+          "description": "One of: SAP_TC2, SAP_CFIN, SAP_APAC, LEGACY_ERP, SHAREPOINT"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entities/entity_type",
+          "name": "entity_type",
+          "dataType": "sc:Text",
+          "description": "One of: product, supplier, tax_code, legal_entity"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entities/name",
+          "name": "name",
+          "dataType": "sc:Text",
+          "description": "System-specific entity name (English, German, abbreviated, or legacy coded)"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "entities/attributes",
+          "name": "attributes",
+          "dataType": "sc:Text",
+          "description": "System-specific attribute dict (schema varies per source system)"
+        }
+      ],
+      "source": {
+        "fileSet": {"@id": "entities-file"},
+        "extract": {"jsonPath": "$[*]"}
+      }
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "pairs",
+      "name": "pairs",
+      "description": "Entity pair judgments with ground truth labels, difficulty tiers, and string similarity scores",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/pair_id",
+          "name": "pair_id",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/entity_a",
+          "name": "entity_a",
+          "dataType": "sc:Text",
+          "references": {"@id": "entities/entity_id"}
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/entity_b",
+          "name": "entity_b",
+          "dataType": "sc:Text",
+          "references": {"@id": "entities/entity_id"}
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/label",
+          "name": "label",
+          "dataType": "sc:Text",
+          "description": "Ground truth: match | no_match | ambiguous"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/difficulty_tier",
+          "name": "difficulty_tier",
+          "dataType": "sc:Text",
+          "description": "One of: easy, medium, hard, adversarial_negative, medium_negative, obvious_negative, ambiguous"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/string_similarity",
+          "name": "string_similarity",
+          "dataType": "sc:Float",
+          "description": "Normalized edit distance similarity between entity_a and entity_b names [0, 1]"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "pairs/entity_type",
+          "name": "entity_type",
+          "dataType": "sc:Text",
+          "description": "Entity type of the pair: product, supplier, tax_code, or legal_entity"
+        }
+      ],
+      "source": {
+        "fileSet": {"@id": "pairs-file"},
+        "extract": {"jsonPath": "$[*]"}
+      }
+    },
+    {
+      "@type": "cr:RecordSet",
+      "@id": "oracle_context",
+      "name": "oracle_context",
+      "description": "Structured oracle context records providing complete canonical cross-system entity mappings",
+      "field": [
+        {
+          "@type": "cr:Field",
+          "@id": "oracle_context/canonical_id",
+          "name": "canonical_id",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "oracle_context/entity_type",
+          "name": "entity_type",
+          "dataType": "sc:Text"
+        },
+        {
+          "@type": "cr:Field",
+          "@id": "oracle_context/system_mappings",
+          "name": "system_mappings",
+          "dataType": "sc:Text",
+          "description": "Dict mapping each source system to the canonical entity's ID and name in that system"
+        }
+      ],
+      "source": {
+        "fileSet": {"@id": "oracle-context-file"},
+        "extract": {"jsonPath": "$[*]"}
+      }
+    }
+  ]
+}