Skip to content

Standardised Files for Text-Based Knowledge Graph Data

A "standardised file" format is used in this project to store knowledge graph (KG) data with rich textual information in a way that is easy to process, share, and load into models. This format ensures consistency across datasets and tools, making it simple to use different KGs with the same codebase.

Why Standardise?

  • Interoperability: Use the same loader and processing code for different datasets (e.g., WN18RR, Wikidata5M).
  • Clarity: Each file has a clear, single purpose (IDs, names, descriptions, relations).
  • Extensibility: Easy to add new datasets or fields.

Standardised File Types

1. entity_ids.txt

A plain text file with one entity ID per line.

Example: 

Q42
Q123
Q999

2. entity_id2_name.json

A JSON dictionary mapping each entity ID to its canonical name.

Example: 

{
  "Q42": "Douglas Adams",
  "Q123": "Example Entity",
  "Q999": "Another Entity"
}

3. entity_id2_description.json

A JSON dictionary mapping each entity ID to a textual description.

Example: 

{
  "Q42": "Douglas Adams was an English author, best known for The Hitchhiker's Guide to the Galaxy.",
  "Q123": "This is a sample entity used for demonstration purposes.",
  "Q999": "A placeholder entity in the knowledge graph."
}

4. relation_id2name.json (or relation_id2description.json)

A JSON dictionary mapping each relation ID to its name or description.

Example: 

{
  "P31": "instance of",
  "P279": "subclass of",
  "P50": "author"
}

Typical Directory Structure

<dataset>_tkg/
    entity_ids.txt
    entity_id2_name.json
    entity_id2_description.json
    relation_id2name.json

Notes

  • All files use UTF-8 encoding.
  • JSON files are formatted for readability (indentation).
  • The same format is used for all supported datasets, enabling unified loading and processing.

Usage

These files can be loaded directly using the provided Python loaders, or processed further for downstream tasks such as training text-based KGC models.