README.md

metadata

task_categories:
  - text-retrieval
  - sentence-similarity
  - text-classification
language:
  - vi
  - zh
tags:
  - IR
  - Rerank
  - Information Retrival
size_categories:
  - 1M<n<10M
configs:
  - config_name: qrels.train
    data_files: qrels.train.tsv
  - config_name: qrels.dev
    data_files:
      - split: train
        path: data_qrels.dev.tsv
  - config_name: qrels.retrieval.train
    data_files:
      - split: train
        path: qrels.retrieval.train.tsv
  - config_name: qrels.retrieval.dev
    data_files:
      - split: train
        path: data_qrels.retrieval.dev.tsv
  - config_name: collection
    data_files:
      - split: train
        path:
          - collection_json_0_vi.tsv
          - collection_json_1_vi.tsv
          - collection_json_2_vi.tsv
          - collection_json_3_vi.tsv
          - collection_json_4_vi.tsv
          - collection_json_5_vi.tsv
          - collection_json_6_vi.tsv
          - collection_json_7_vi.tsv
          - collection_json_8_vi.tsv
          - collection_json_9_vi.tsv
  - config_name: queries.train
    data_files:
      - split: train
        path:
          - data_queries.train_json_vi.tsv
  - config_name: queries.dev
    data_files:
      - split: train
        path:
          - data_queries.dev_json_vi.tsv
  - config_name: queries.test
    data_files:
      - split: train
        path:
          - data_queries.test_json_vi.tsv
  - config_name: train.bm25.tsv
    data_files:
      - split: train
        path:
          - train.bm25.tsv
  - config_name: train.mined.tsv
    data_files:
      - split: train
        path:
          - train.mined.tsv
license: apache-2.0

📚 5CD-AI/Vietnamese-THUIR-T2Ranking-gg-translated

📝 Overview

Vietnamese-THUIR-T2Ranking-gg-translated is a large-scale dataset for passage ranking in Vietnamese.
It is translated from the original THUIR/T2Ranking [1] using Google Translate, inspired by the approach of mMARCO [2].
The dataset aims to provide a large-scale dataset for research and applications in Information Retrieval (IR) in Vietnamese.

In IR, passage ranking is an essential and challenging task, typically involving two stages:

1. 🔍 Passage retrieval – retrieving candidate passages.
2. 📊 Passage re-ranking – re-ordering candidates for final ranking.

📥 Data Download

Data files follow the same structure as the original:

📂 Description	🗂️ Filename	🔢 Num Records	📑 Format
Collection	collection_json_*_vi.tsv	2,303,643	tsv: pid, text_zh, text_vi
Queries Train	data_queries.train_json_vi.tsv	258,042	tsv: qid, text_zh, text_vi
Queries Dev	data_queries.dev_json_vi.tsv	24,832	tsv: qid, text_zh, text_vi
Queries Test	data_queries.test_json_vi.tsv	24,832	tsv: qid, text_zh, text_vi
Qrels Train for re-ranking	qrels.train.tsv	1,613,421	TREC qrels format
Qrels Dev for re-ranking	qrels.dev.tsv	400,536	TREC qrels format
Qrels Retrieval Train	qrels.retrieval.train.tsv	744,663	tsv: qid, pid
Qrels Retrieval Dev	qrels.retrieval.dev.tsv	118,933	tsv: qid, pid
BM25 Negatives	train.bm25.tsv	200,359,731	tsv: qid, pid, index
Hard Negatives	train.mined.tsv	200,376,001	tsv: qid, pid, index, score

🚀 How to download

git lfs install
git clone https://huggingface.co/datasets/5CD-AI/Vietnamese-THUIR-T2Ranking-gg-translated

📂 Folder structure:

├── collection_json_*_vi.tsv
├── data_queries.train_json_vi.tsv
├── data_queries.dev_json_vi.tsv
├── data_queries.test_json_vi.tsv
├── qrels.train.tsv
├── qrels.dev.tsv
├── qrels.retrieval.train.tsv
├── qrels.retrieval.dev.tsv
├── train.bm25.tsv
└── train.mined.tsv

🗒️ Notes

• ⚠️ This dataset was translated using Google Translate, so some translations may be imperfect or unnatural.

📖 Reference

[1] X. Xie et al., T2Ranking: A Large-scale Chinese Benchmark for Passage Ranking, in Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’23), ACM, pp. 2681–2690, 2023. doi: 10.1145/3539618.3591874.

[2] L. H. Bonifacio et al., mMARCO: A Multilingual Version of MS MARCO Passage Ranking Dataset, arXiv preprint arXiv:2108.13897, 2021.