← home
Github: datasets/neuclir.py

ir_datasets: NeuCLIR Corpus

Index
  1. neuclir
  2. neuclir/1
  3. neuclir/1/fa
  4. neuclir/1/fa/hc4-filtered
  5. neuclir/1/ru
  6. neuclir/1/ru/hc4-filtered
  7. neuclir/1/zh
  8. neuclir/1/zh/hc4-filtered

Data Access Information

To access the docuemnts of this dataset, you will need to download the documents from Common Crawl. The script for downloading and validating the documents are in NeuCLIR/download-collection . Please use the following command to download the documents:

git clone https://github.com/NeuCLIR/download-collection
cd download-collection
pip install -r requirements.txt
python download_documents.py --storage ~/.ir_datasets/neuclir/1 \
--zho ./resource/zho/ids.jsonl.gz \
--fas ./resource/fas/ids.jsonl.gz \
--rus ./resource/rus/ids.*.jsonl.gz \
--jobs {number of process}

After download, please also post-process the downloaded file to verify all and only specified documents are downloaded, and modify the ordering of the collection to match the original specified ordering in the id files.

for lang in zho fas rus; do
  python fix_document_order.py --raw_download_file ~/.ir_datasets/neuclir/1/$lang/docs.jsonl \
   --id_file ./resource/$lang/ids*.jsonl.gz \
   --check_hash
done

You can also store the documents in other directory and create a soft link for ~/.ir_datasets/neuclir/22/.

"neuclir"

This is the dataset created for TREC 2022 NeuCLIR Track. Topics will be developed and released by June 2022 by NIST. Relevance judgements will be available after the evaluation (around November).

The collection designed to be similar to [HC4] and a large portion of documents from HC4 are ported to this collection. Users can conduct experiemnts on this collection with queries and qrels in HC4 for development.

  • Documents: Web pages from Common Crawl in Chinese, Persian, and Russian.
  • Queries: (To be released) English TREC-style title/description queries. Narrative field contains an example passage for each relevance level. Human and machine translation of the titles and descriptions in the target language (i.e., document language) are provided in the query object.
  • Qrels: (To be released) Documents are judged in three levels of relevance. Please refer to the dataset paper for the full definition of the levels.
  • See also: hc4
  • NeuCLIR Track Website
  • Collection Repository

"neuclir/1"

Version 1 of the NeuCLIR corpus.

"neuclir/1/fa"

The Persian collection contains English queries (to be released) and Persian documents for retrieval. Human and machine translated queries will be provided in the query object for running monolingual retrieval or cross-language retrival assuming the machine query tranlstion into Persian is available.

docs
2.2M docs

Language: fa

Document type:
ExctractedCCDoc: (namedtuple)
  1. doc_id: str
  2. title: str
  3. text: str
  4. url: str
  5. time: str
  6. cc_file: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/fa")
for doc in dataset.docs_iter():
    doc # namedtuple<doc_id, title, text, url, time, cc_file>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/fa docs
[doc_id]    [title]    [text]    [url]    [time]    [cc_file]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

Metadata 
{
  "docs": {
    "count": 2232016,
    "fields": {
      "doc_id": {
        "max_len": 36,
        "common_prefix": ""
      }
    }
  }
}

"neuclir/1/fa/hc4-filtered"

Subset of the Persian collection that intersect with HC4. The 60 queries are the hc4/fa/dev and hc4/fa/test sets combined.

queries
60 queries

Language: multiple/other/unknown

Query type:
ExctractedCCQuery: (namedtuple)
  1. query_id: str
  2. title: str
  3. description: str
  4. ht_title: str
  5. ht_description: str
  6. mt_title: str
  7. mt_description: str
  8. narrative_by_relevance: Dict[str,str]
  9. report: str
  10. report_url: str
  11. report_date: str
  12. translation_lang: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/fa/hc4-filtered")
for query in dataset.queries_iter():
    query # namedtuple<query_id, title, description, ht_title, ht_description, mt_title, mt_description, narrative_by_relevance, report, report_url, report_date, translation_lang>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/fa/hc4-filtered queries
[query_id]    [title]    [description]    [ht_title]    [ht_description]    [mt_title]    [mt_description]    [narrative_by_relevance]    [report]    [report_url]    [report_date]    [translation_lang]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

docs
392K docs

Language: fa

Document type:
ExctractedCCDoc: (namedtuple)
  1. doc_id: str
  2. title: str
  3. text: str
  4. url: str
  5. time: str
  6. cc_file: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/fa/hc4-filtered")
for doc in dataset.docs_iter():
    doc # namedtuple<doc_id, title, text, url, time, cc_file>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/fa/hc4-filtered docs
[doc_id]    [title]    [text]    [url]    [time]    [cc_file]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

qrels
3.1K qrels
Query relevance judgment type:
TrecQrel: (namedtuple)
  1. query_id: str
  2. doc_id: str
  3. relevance: int
  4. iteration: str

Relevance levels

Rel.DefinitionCount%
0Not-valuable. Information in the document might be included in a report footnote, or omitted entirely.2.6K82.8%
1Somewhat-valuable. The most valuable information in the document would be found in the remainder of such a report.261 8.5%
3Very-valuable. Information in the document would be found in the lead paragraph of a report that is later written on the topic.269 8.7%

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/fa/hc4-filtered")
for qrel in dataset.qrels_iter():
    qrel # namedtuple<query_id, doc_id, relevance, iteration>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/fa/hc4-filtered qrels --format tsv
[query_id]    [doc_id]    [relevance]    [iteration]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

Citation

ir_datasets.bib:

\cite{Lawrie2022HC4}

Bibtex:

@article{Lawrie2022HC4, author = {Dawn Lawrie and James Mayfield and Douglas W. Oard and Eugene Yang}, title = {HC4: A New Suite of Test Collections for Ad Hoc CLIR}, booktitle = {{Advances in Information Retrieval. 44th European Conference on IR Research (ECIR 2022)}, year = {2022}, month = apr, publisher = {Springer}, series = {Lecture Notes in Computer Science}, site = {Stavanger, Norway}, url = {https://arxiv.org/abs/2201.09992} }
Metadata 
{
  "docs": {
    "count": 391703,
    "fields": {
      "doc_id": {
        "max_len": 36,
        "common_prefix": ""
      }
    }
  },
  "queries": {
    "count": 60
  },
  "qrels": {
    "count": 3087,
    "fields": {
      "relevance": {
        "counts_by_value": {
          "0": 2557,
          "3": 269,
          "1": 261
        }
      }
    }
  }
}

"neuclir/1/ru"

The Russian collection contains English queries (to be released) and Russian documents for retrieval. Human and machine translated queries will be provided in the query object for running monolingual retrieval or cross-language retrival assuming the machine query tranlstion into Russian is available.

docs
4.6M docs

Language: ru

Document type:
ExctractedCCDoc: (namedtuple)
  1. doc_id: str
  2. title: str
  3. text: str
  4. url: str
  5. time: str
  6. cc_file: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/ru")
for doc in dataset.docs_iter():
    doc # namedtuple<doc_id, title, text, url, time, cc_file>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/ru docs
[doc_id]    [title]    [text]    [url]    [time]    [cc_file]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

Metadata 
{
  "docs": {
    "count": 4627543,
    "fields": {
      "doc_id": {
        "max_len": 36,
        "common_prefix": ""
      }
    }
  }
}

"neuclir/1/ru/hc4-filtered"

Subset of the Russian collection that intersect with HC4. The 54 queries are the hc4/ru/dev and hc4/ru/test sets combined.

queries
54 queries

Language: multiple/other/unknown

Query type:
ExctractedCCQuery: (namedtuple)
  1. query_id: str
  2. title: str
  3. description: str
  4. ht_title: str
  5. ht_description: str
  6. mt_title: str
  7. mt_description: str
  8. narrative_by_relevance: Dict[str,str]
  9. report: str
  10. report_url: str
  11. report_date: str
  12. translation_lang: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/ru/hc4-filtered")
for query in dataset.queries_iter():
    query # namedtuple<query_id, title, description, ht_title, ht_description, mt_title, mt_description, narrative_by_relevance, report, report_url, report_date, translation_lang>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/ru/hc4-filtered queries
[query_id]    [title]    [description]    [ht_title]    [ht_description]    [mt_title]    [mt_description]    [narrative_by_relevance]    [report]    [report_url]    [report_date]    [translation_lang]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

docs
965K docs

Language: ru

Document type:
ExctractedCCDoc: (namedtuple)
  1. doc_id: str
  2. title: str
  3. text: str
  4. url: str
  5. time: str
  6. cc_file: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/ru/hc4-filtered")
for doc in dataset.docs_iter():
    doc # namedtuple<doc_id, title, text, url, time, cc_file>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/ru/hc4-filtered docs
[doc_id]    [title]    [text]    [url]    [time]    [cc_file]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

qrels
3.2K qrels
Query relevance judgment type:
TrecQrel: (namedtuple)
  1. query_id: str
  2. doc_id: str
  3. relevance: int
  4. iteration: str

Relevance levels

Rel.DefinitionCount%
0Not-valuable. Information in the document might be included in a report footnote, or omitted entirely.2.5K76.8%
1Somewhat-valuable. The most valuable information in the document would be found in the remainder of such a report.478 14.8%
3Very-valuable. Information in the document would be found in the lead paragraph of a report that is later written on the topic.274 8.5%

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/ru/hc4-filtered")
for qrel in dataset.qrels_iter():
    qrel # namedtuple<query_id, doc_id, relevance, iteration>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/ru/hc4-filtered qrels --format tsv
[query_id]    [doc_id]    [relevance]    [iteration]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

Citation

ir_datasets.bib:

\cite{Lawrie2022HC4}

Bibtex:

@article{Lawrie2022HC4, author = {Dawn Lawrie and James Mayfield and Douglas W. Oard and Eugene Yang}, title = {HC4: A New Suite of Test Collections for Ad Hoc CLIR}, booktitle = {{Advances in Information Retrieval. 44th European Conference on IR Research (ECIR 2022)}, year = {2022}, month = apr, publisher = {Springer}, series = {Lecture Notes in Computer Science}, site = {Stavanger, Norway}, url = {https://arxiv.org/abs/2201.09992} }
Metadata 
{
  "docs": {
    "count": 964719,
    "fields": {
      "doc_id": {
        "max_len": 36,
        "common_prefix": ""
      }
    }
  },
  "queries": {
    "count": 54
  },
  "qrels": {
    "count": 3235,
    "fields": {
      "relevance": {
        "counts_by_value": {
          "0": 2483,
          "1": 478,
          "3": 274
        }
      }
    }
  }
}

"neuclir/1/zh"

The Chinese collection contains English queries (to be released) and Chinese documents for retrieval. Human and machine translated queries will be provided in the query object for running monolingual retrieval or cross-language retrival assuming the machine query tranlstion into Chinese is available.

docs
3.2M docs

Language: zh

Document type:
ExctractedCCDoc: (namedtuple)
  1. doc_id: str
  2. title: str
  3. text: str
  4. url: str
  5. time: str
  6. cc_file: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/zh")
for doc in dataset.docs_iter():
    doc # namedtuple<doc_id, title, text, url, time, cc_file>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/zh docs
[doc_id]    [title]    [text]    [url]    [time]    [cc_file]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

Metadata 
{
  "docs": {
    "count": 3179209,
    "fields": {
      "doc_id": {
        "max_len": 36,
        "common_prefix": ""
      }
    }
  }
}

"neuclir/1/zh/hc4-filtered"

Subset of the Chinse collection that intersect with HC4. The 60 queries are the hc4/zh/dev and hc4/zh/test sets combined.

queries
60 queries

Language: multiple/other/unknown

Query type:
ExctractedCCQuery: (namedtuple)
  1. query_id: str
  2. title: str
  3. description: str
  4. ht_title: str
  5. ht_description: str
  6. mt_title: str
  7. mt_description: str
  8. narrative_by_relevance: Dict[str,str]
  9. report: str
  10. report_url: str
  11. report_date: str
  12. translation_lang: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/zh/hc4-filtered")
for query in dataset.queries_iter():
    query # namedtuple<query_id, title, description, ht_title, ht_description, mt_title, mt_description, narrative_by_relevance, report, report_url, report_date, translation_lang>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/zh/hc4-filtered queries
[query_id]    [title]    [description]    [ht_title]    [ht_description]    [mt_title]    [mt_description]    [narrative_by_relevance]    [report]    [report_url]    [report_date]    [translation_lang]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

docs
520K docs

Language: zh

Document type:
ExctractedCCDoc: (namedtuple)
  1. doc_id: str
  2. title: str
  3. text: str
  4. url: str
  5. time: str
  6. cc_file: str

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/zh/hc4-filtered")
for doc in dataset.docs_iter():
    doc # namedtuple<doc_id, title, text, url, time, cc_file>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/zh/hc4-filtered docs
[doc_id]    [title]    [text]    [url]    [time]    [cc_file]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

qrels
3.2K qrels
Query relevance judgment type:
TrecQrel: (namedtuple)
  1. query_id: str
  2. doc_id: str
  3. relevance: int
  4. iteration: str

Relevance levels

Rel.DefinitionCount%
0Not-valuable. Information in the document might be included in a report footnote, or omitted entirely.2.7K82.4%
1Somewhat-valuable. The most valuable information in the document would be found in the remainder of such a report.222 6.9%
3Very-valuable. Information in the document would be found in the lead paragraph of a report that is later written on the topic.344 10.7%

Examples:

Python API 
import ir_datasets
dataset = ir_datasets.load("neuclir/1/zh/hc4-filtered")
for qrel in dataset.qrels_iter():
    qrel # namedtuple<query_id, doc_id, relevance, iteration>

You can find more details about the Python API here.

CLI 
ir_datasets export neuclir/1/zh/hc4-filtered qrels --format tsv
[query_id]    [doc_id]    [relevance]    [iteration]
...

You can find more details about the CLI here.

PyTerrier

No example available for PyTerrier

Citation

ir_datasets.bib:

\cite{Lawrie2022HC4}

Bibtex:

@article{Lawrie2022HC4, author = {Dawn Lawrie and James Mayfield and Douglas W. Oard and Eugene Yang}, title = {HC4: A New Suite of Test Collections for Ad Hoc CLIR}, booktitle = {{Advances in Information Retrieval. 44th European Conference on IR Research (ECIR 2022)}, year = {2022}, month = apr, publisher = {Springer}, series = {Lecture Notes in Computer Science}, site = {Stavanger, Norway}, url = {https://arxiv.org/abs/2201.09992} }
Metadata 
{
  "docs": {
    "count": 519945,
    "fields": {
      "doc_id": {
        "max_len": 36,
        "common_prefix": ""
      }
    }
  },
  "queries": {
    "count": 60
  },
  "qrels": {
    "count": 3217,
    "fields": {
      "relevance": {
        "counts_by_value": {
          "0": 2651,
          "3": 344,
          "1": 222
        }
      }
    }
  }
}