Skip to content

Dataset · mteb

Counter-argument retrieval benchmark for embeddings (ArguAna)

mteb/arguana

A set of text pairs from the MTEB benchmark to evaluate the quality of finding the best counter-argument without topic knowledge.

What's inside

The dataset is part of the well-known MTEB embedding benchmark and is a corpus of English texts ranging from 10 to 100 thousand records. The data is taken from social, web, and written sources and is intended for the text-to-text retrieval task of finding a suitable counter-argument. The card provides a ready-made example of evaluating models via the mteb library.

What you can do with it

  • Evaluate and compare text embedding models on semantic search tasks
  • Test RAG systems and algorithms for searching relevant arguments in discussions
Tasks
Text retrieval
Languages
English
Size
10K–100K rows
Downloads this month
18.9K
Likes
13

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("mteb/arguana")