README.md

metadata

language:
  - en
multilinguality:
  - monolingual
task_categories:
  - text-retrieval
source_datasets:
  - NevIR
task_ids:
  - document-retrieval
config_names:
  - corpus
  - queries
  - qrels
  - top_ranked
tags:
  - text-retrieval
  - negation
dataset_info:
  - config_name: corpus
    features:
      - name: _id
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: corpus
        num_examples: 1896
  - config_name: queries
    features:
      - name: _id
        dtype: string
      - name: text
        dtype: string
    splits:
      - name: queries
        num_examples: 1896
  - config_name: qrels
    features:
      - name: query_id
        dtype: string
      - name: doc_id
        dtype: string
      - name: score
        dtype: int
    splits:
      - name: train
        num_examples: 1896
      - name: validation
        num_examples: 450
      - name: test
        num_examples: 450
  - config_name: top_ranked
    features:
      - name: query_id
        dtype: string
      - name: doc_ids
        dtype: list
    splits:
      - name: train
        num_examples: 1896
      - name: validation
        num_examples: 238
      - name: test
        num_examples: 238
configs:
  - config_name: corpus
    data_files:
      - split: corpus
        path: corpus.jsonl
  - config_name: queries
    data_files:
      - split: queries
        path: queries.jsonl
  - config_name: qrels
    data_files:
      - split: train
        path: qrels/train.jsonl
      - split: validation
        path: qrels/validation.jsonl
      - split: test
        path: qrels/test.jsonl
  - config_name: top_ranked
    data_files:
      - split: train
        path: top_ranked/train.jsonl
      - split: validation
        path: top_ranked/validation.jsonl
      - split: test
        path: top_ranked/test.jsonl

NevIR-mteb Dataset

This is the MTEB-compatible version of the NevIR dataset, structured for information retrieval tasks focused on negation understanding.

Dataset Structure

The dataset is organized into multiple configurations:

1. corpus: Contains all documents (doc1 and doc2 from each sample)
2. queries: Contains all queries (q1 and q2 from each sample)
3. qrels: Contains relevance judgments (q1 matches with doc1, q2 matches with doc2)
4. top_ranked: Contains candidate documents for each query (both doc1 and doc2 for every query)

Usage

from datasets import load_dataset

# Load the entire dataset
dataset = load_dataset("orionweller/NevIR-mteb")

# Load specific configurations
corpus = load_dataset("orionweller/NevIR-mteb", "corpus")
queries = load_dataset("orionweller/NevIR-mteb", "queries")
qrels = load_dataset("orionweller/NevIR-mteb", "qrels")
top_ranked = load_dataset("orionweller/NevIR-mteb", "top_ranked")