Skip to content

Dataset · philippesaade

Wikidata entity and property vector embeddings

philippesaade/Wikidata_Vectors_0.2

Tens of millions of vectors for Wikidata elements in 4 languages from Wikimedia, Jina AI, and IBM.

What's inside

The dataset contains between 10 million and 100 million vector representations (embeddings) for entities (Q...) and properties (P...) of the Wikidata knowledge base, built based on their textual descriptions. English, Arabic, French, and German languages are supported. The project was created by Wikimedia Deutschland in cooperation with Jina AI and IBM DataStax to develop public vector knowledge bases.

What you can do with it

  • Building semantic search and RAG systems on top of the Wikidata knowledge graph
  • Training and evaluating entity linking models in texts
  • Feature extraction and clustering of multilingual concepts
Tasks
Embeddings, Text retrieval, Text classification, Sentence similarity
Languages
English, Arabic, French, German
Size
10M–100M rows
Downloads this month
20.6K
Likes
4

How to load it

from datasets import load_dataset

ds = load_dataset("philippesaade/Wikidata_Vectors_0.2")