Omnivent

Dataset description

Omnivent is a corpus of sentences annotated for the event extraction task by a Large Language Model (LLM). More precisely, these sentences were drawn from a subset of the CC-News dataset comprising English-language news articles published between January 2017 and December 2019. Each sentence is annotated for at least one event. The annotation of an event covers both its mention in the sentence and the mentions of its arguments, each mention being associated with its string and its type.

Statistics

	Number of elements
Documents	58,064
Sentences	101000
Event types	38,859
Event mentions	199,716
Argument roles	9,981
Arguments	522,703

Dataset Structure

Dataset instances

Each element of the file is a sentence with the annotation of the events for which an annotation was produced. For each annotated event, both the event trigger and the event arguments are given.

  {
    "sentence": "After the third missed payment, you should receive a Demand Letter notifying you of your delinquency",
    "doc_id": 5,
    "index": 5,
    "isEvent": {
      "flag": true,
      "prediction confidence": 0.9
    },
    "annotation": {
      "events": [
        {
          "event trigger": "receive",
          "event categories": [
            "Communication:Receipt"
          ],
          "arguments": [
            {
              "entity": "you",
              "entity type": [
                "Recipient"
              ]
            },
            {
              "entity": "Demand Letter",
              "entity type": [
                "Document"
              ]
            },
            {
              "entity": "delinquency",
              "entity type": [
                "Reason"
              ]
            }
          ],
          "prediction confidence": 0.9
        }
      ]
    }
  }

Data fields

• sentence (str): sentence string
• doc_id (int): identifier of the document from which the sentence comes from
• isEvent.prediction confidence (float): confidence score of the LLM about the presence of an event in the sentence
• events ([]]): list of events annotated in the sentence
• event trigger (str): string of an event mention
• event categories ([str]]): type of an event mention
• arguments ([]]): list of arguments associated with an event
• entity (str): string of an event argument mention
• entity type ([str]): type of an event argument mention
• events[].prediction confidence (float): confidence score of the LLM about the annotation of an annotation in the sentence

Citation

@online{omnivent_dataset,
    author = "Salim Abdou Daoura and Sondes Bannour Souihi and Romaric Besançon and Olivier Ferret",
    title  = "Omnivent",
    url    = "https://huggingface.co/datasets/cea-list-ia/omnivent"
}