ragtime1 / README.md
eugene-yang's picture
Update README.md
32d7756 verified
metadata
license: apache-2.0
task_categories:
  - text-retrieval
task_ids:
  - document-retrieval
language:
  - ar
  - en
  - ru
  - zh
multilinguality:
  - multilingual
tags:
  - multilingual
  - RAG
  - News
source_datasets:
  - extended|c4
annotations_creators:
  - no-annotation
size_categories:
  - 1M<n<10M
pretty_name: RAGTIME1
configs:
  - config_name: default
    data_files:
      - split: arb
        path: arb-docs.jsonl
      - split: rus
        path: rus-docs.jsonl
      - split: zho
        path: zho-docs.jsonl
      - split: eng
        path: eng-docs.jsonl
  - config_name: mt_docs
    data_files:
      - split: arb.mt.eng
        path: arb-trans.jsonl
      - split: rus.mt.eng
        path: rus-trans.jsonl
      - split: zho.mt.eng
        path: zho-trans.jsonl
dataset_info:
  - config_name: default
    dataset_info:
      splits:
        - name: arb
          num_bytes: 3480001855
          num_examples: 1000095
        - name: arb.mt.eng
          num_bytes: 3382078786
          num_examples: 1000095
        - name: rus
          num_bytes: 3196832604
          num_examples: 1000095
  - config_name: mt_docs
    dataset_info:
      splits:
        - name: rus.mt.eng
          num_bytes: 2640657225
          num_examples: 1000095
        - name: zho
          num_bytes: 2541033536
          num_examples: 1000095
        - name: zho.mt.eng
          num_bytes: 3787303609
          num_examples: 1000095
        - name: eng
          num_bytes: 3399011290
          num_examples: 1000095

RAGTIME1 Collection

This dataset contains the documents for TREC RAGTIME Track. Please refer to the website for the details of the task.

RAGTIME is a multilingual RAG task, which expects the participating system to retrieve relevant documents from all four languages and synthesize a response with citation to the report request. For convenience, we separate the documents by their languages into four .jsonl files. However, they are intended to be used as a whole set.

The documents are extracted from Common Crawl News and sampled between August 1, 2021, and July 31, 2024, with an even number of documents every day. Each language has 1,000,095 documents. Machine translation using a Sockeye model trained by HLTCOE are also released.

TREC RAGTIME Track 2025 Timeline

TREC RAGTIME Track 2025 has concluded. Thank you all for participating.

We will be back in TREC 2026 and please join us again.

If you are interested in more report generation, please come join us at RAG4Report @ ACL2026, which features an additional automatic report evaluation shared task and more!

Citation and Attribution

If you use this collection, please cite our overview paper, which does not exist yet.

The best way to provide attribution is to participate in our shared task. We would love to have your RAG systems and evaluate them :)