dataset.croissant.json

{
  "@context": {
    "@language": "en",
    "@vocab": "https://schema.org/",
    "sc": "https://schema.org/",
    "cr": "http://mlcommons.org/croissant/",
    "rai": "http://mlcommons.org/croissant/RAI/",
    "dct": "http://purl.org/dc/terms/",
    "prov": "http://www.w3.org/ns/prov#",

    "includes": "http://mlcommons.org/schema/includes",
    "extract": "cr:extract",
    "fileSet": "cr:fileSet",
    "field": "cr:field",
    "dataType": "cr:dataType",
    "source": "cr:source",
    "jsonPath": "cr:jsonPath",
    "recordSet": "cr:recordSet"
  },

  "@type": "sc:Dataset",
  "name": "ViMU",
  "description": "ViMU is a benchmark for video metaphorical understanding, designed to evaluate whether video-language models can infer implicit, subtextual, rhetorical, and socially situated meanings beyond literal visual perception.",
  "url": "https://huggingface.co/datasets/YOUR_ANON/ViMU",
  "license": "https://creativecommons.org/licenses/by/4.0/",
  "conformsTo": "http://mlcommons.org/croissant/1.1",

  "citation": "Anonymous. ViMU: Benchmarking Video Metaphorical Understanding. Under review, 2026.",
  "datePublished": "2026-05-05",
  "version": "1.0.0",

  "distribution": [
    {
      "@type": "sc:FileSet",
      "@id": "videos",
      "name": "videos",
      "description": "Video files used in the ViMU benchmark.",
      "encodingFormat": "video/mp4",
      "includes": "videos/*.mp4"
    },
    {
      "@type": "sc:FileSet",
      "@id": "metadata",
      "name": "metadata",
      "description": "JSONL metadata files containing video identifiers, questions, answers, taxonomy annotations, and evaluation metadata.",
      "encodingFormat": "application/jsonlines",
      "includes": "metadata/*.jsonl"
    }
  ],

  "recordSet": [
    {
      "@type": "cr:RecordSet",
      "@id": "OpenEndedQA",
      "name": "OpenEndedQA",
      "description": "Open-ended question-answer records for evaluating video metaphorical understanding.",
      "field": [
        {
          "@type": "cr:Field",
          "@id": "OpenEndedQA/video_id",
          "name": "video_id",
          "description": "Unique identifier of the video.",
          "dataType": "sc:Text",
          "source": {
            "fileSet": {
              "@id": "metadata"
            },
            "extract": {
              "jsonPath": "$.video_id"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "OpenEndedQA/question",
          "name": "question",
          "description": "Open-ended question associated with the video.",
          "dataType": "sc:Text",
          "source": {
            "fileSet": {
              "@id": "metadata"
            },
            "extract": {
              "jsonPath": "$.qa.question"
            }
          }
        },
        {
          "@type": "cr:Field",
          "@id": "OpenEndedQA/answer",
          "name": "answer",
          "description": "Reference answer for the open-ended question.",
          "dataType": "sc:Text",
          "source": {
            "fileSet": {
              "@id": "metadata"
            },
            "extract": {
              "jsonPath": "$.qa.answer"
            }
          }
        }
      ]
    }
  ],

  "rai:dataLimitations": "ViMU is intended for evaluating video metaphorical and subtextual understanding. It should not be interpreted as a complete measure of general video intelligence, cultural competence, or social reasoning. Performance may depend on cultural background knowledge, platform-specific conventions, and the model's ability to process videos.",
  "rai:dataBiases": "The dataset is curated from videos and may overrepresent internet-native, English-facing, or platform-specific humor and rhetoric. Some social groups, cultural contexts, rhetorical styles, or meme formats may be underrepresented. These biases may affect model performance across cultural and demographic contexts.",
  "rai:personalSensitiveInformation": "The dataset may contain publicly visible people, social signals, political or identity-related references, and socially sensitive implications present in videos. The benchmark is designed for evaluation and analysis, not for identifying individuals or inferring private attributes.",
  "rai:dataUseCases": [
    "Evaluation of video-language models on metaphorical, implicit, and subtextual understanding.",
    "Analysis of model robustness on rhetoric, social value signals, and evidence grounding.",
    "Benchmarking model behavior beyond literal visual perception."
  ],
  "rai:dataSocialImpact": "ViMU can help identify limitations in video-language models' understanding of implicit meanings, social cues, and rhetorical mechanisms. Potential risks include misuse for profiling, moderation automation without human oversight, or overgeneralized claims about social reasoning. The dataset should be used with careful reporting of limitations and error modes.",
  "rai:hasSyntheticData": false,
  "prov:wasDerivedFrom": [
    {
      "@type": "prov:Entity",
      "name": "Publicly available online videos",
      "description": "The dataset was derived from publicly available online videos that were curated, filtered, annotated, and reorganized for research and benchmark evaluation purposes. The released dataset provides benchmark metadata and task annotations for evaluating video metaphorical understanding."
    }
  ],
  "prov:wasGeneratedBy": [
    {
      "@type": "prov:Activity",
      "name": "Data collection and curation",
      "description": "Videos were curated for benchmark construction, with associated metadata, questions, answers, evidence annotations, and taxonomy labels prepared for evaluating metaphorical video understanding."
    },
    {
      "@type": "prov:Activity",
      "name": "Annotation and validation",
      "description": "Annotations include open-ended QA, evidence grounding labels, rhetorical mechanism labels, social value signal labels, and validation or filtering steps designed to improve benchmark quality."
    }
  ]
}