Low_Resource_KBP

knowledge graph population in low resource conditions

The file "Few-Shot_ED.json.zip" is the FewEvent dataset for the paper accepted by WSDM 2020 "Meta-Learning with Dynamic-Memory-Based Prototypical Network for Few-Shot Event Detection"

Source of Raw Data

• We first scale up the number of event types in existing datasets, including the ACE-2005 corpus, and TAC-KBP-2017 Event Track Data.

• We then import and extend some new event types based on an automatically-labeled event data, from Freebase and Wikipedia, constrained to specific domains such as music, film, sports, education, etc.

Data Structure

In "Few-Shot_ED.json.zip"，the key is "event type label", the value is the event instances.

{event_type_label1: [instance1, instance2, ...], event_type_label2: ...}

example of an event instance:

(Note that the sentence should be cleaned, e.g., To expand abbreviations)

["In trucks and on foot they came to the town of Safwan", "came", [7, 6, 12]]

• "In trucks and on foot they came to the town of Safwan" is an event mention
• "came" is trigger
• 7 means the distance from trigger "came" to the beginning of the sentence
• 6 means the distance from trigger "came" to the ending of the sentence
• 12 means the sentence length