README.md

---
task_categories:
- text-retrieval
language:
- ja
- en
tags:
- ntcir
pretty_name: NTCIR Data Search Task Test Collection
viewer: false
---
# NTCIR Data Search Task Test Collection

## License

The contents of the following files are distributed under different licenses.
Please refer to `attribution` field in `data_search_*_collection.jsonl.bz2`
for details.
- `data_search_j_collection.jsonl.bz2`
- `data_search_e_collection.jsonl.bz2`
- `data_search_j_data.tar.bz2`
- `data_search_e_data.tar.bz2`

The other files are licensed under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).
Please cite **our papers** presented at https://ntcir.datasearch.jp/ , and do not cite our website.

## Collections

- `data_search_j_collection.jsonl.bz2` (Japanese subtask)
- `data_search_e_collection.jsonl.bz2` (English subtask)
	- Each line of these files is a dataset in the collection, and is in the JSON format
	- Each dataset may contain multiple data, which are in `data` field
	- Each data is stored in a file whose name is described in `data_filename` field

---

- `data_search_j_data.tar.bz2` (Japanese subtask)
- `data_search_e_data.tar.bz2` (English subtask)
	- All the data in the datasets are included in these files
	- These files are NOT included in Google Drive. Please visit https://ntcir.datasearch.jp/data/ to download the data files. 

---

- `checksums.txt`
	- MD5 hash values of `bz2` files are included



## Training queries for NTCIR-15 Data Search

- `data_search_j_train_topics.tsv` (Japanese subtask)
- `data_search_e_train_topics.tsv` (English subtask)
	- Training queries, by which participants are expected to develop systems.
	- Each line contains an ID and a query string separated by TAB


## Test queries for NTCIR-15 Data Search

- `data_search_j_test_topics.tsv` (Japanese subtask)
- `data_search_e_test_topics.tsv` (English subtask)


## Training qrels for NTCIR-15 Data Search

Relevance judgments for training queries.

- `data_search_j_train_qrels.txt` (Japanese subtask)
- `data_search_e_train_qrels.txt` (English subtask)
	- The same format as that used in the NTCIR WWW tasks
		- `[QUERY_ID] [DOC_ID] [REL_LEVEL]`
		- e.g. `DS1-J-0002 000031519435 L2`, meaning that dataset `000031519435` is highly relevant (`L2`) for query `DS1-J-0002`.


## Test qrels for NTCIR-15 Data Search

Relevance judgments for test queries.
Documents from the official runs were judged.

- `data_search_j_test_qrels.txt` (Japanese subtask)
- `data_search_e_test_qrels.txt` (English subtask)


## Extra Qrels for NTCIR-15 Data Search

Relevance judgments for test queries.
Documents from the extra runs were judged.

- `data_search_j_extra_qrels.txt` (Japanese subtask)
- `data_search_e_extra_qrels.txt` (English subtask)


## All Qrels for NTCIR-15 Data Search

These are sets of qrels including the test qrels and extra qrels.

- `data_search_j_all_qrels.txt` (Japanese subtask)
- `data_search_e_all_qrels.txt` (English subtask)


## Training queries for NTCIR-16 Data Search

Training and test queries used in NTCIR-15 Data Search,
which could be used as training queries in NTCIR-16 Data Search 2.
More precisely, `data_search_2_*_train_topics.tsv` is a union of
`data_search_*_train_topics.tsv` and `data_search_*_test_topics.tsv`.

- `data_search_2_j_train_topics.tsv` (Japanese subtask)
- `data_search_2_e_train_topics.tsv` (English subtask)


## Training qrels for NTCIR-16 Data Search

Qrels for training and test queries in NTCIR-15 Data Search.
More precisely, `data_search_2_*_train_qrels.txt` is a union of
`data_search_*_train_qrels.txt` and `data_search_*_all_qrels.txt`.

- `data_search_2_j_train_qrels.txt` (Japanese subtask)
- `data_search_2_e_train_qrels.txt` (English subtask)


## Test queries for NTCIR-16 Data Search 2

- `data_search_2_j_test_topics.tsv` (Japanese subtask)
- `data_search_2_e_test_topics.tsv` (English subtask)


## Test qrels for NTCIR-16 Data Search 2

- `data_search_2_j_test_qrels.txt` (Japanese subtask)
- `data_search_2_e_test_qrels.txt` (English subtask)