uva-irlab/trec-cast-2019-multi-turn

---
language:
- en
multilinguality:
- monolingual
size_categories:
- 10M<n<100M
task_categories:
- text-retrieval
task_ids:
- document-retrieval
language_bcp47:
- en-US
---

# TREC Cast 2019

[TREC Cast](http://www.treccast.ai) have released a document collection with topics and qrels of which a subset has been annotated such that it is suitable for multi-turn conversational search.

## Dataset statistics

- # Passages: 38,426,252
- # Topics: 20
- # Queries: 173

## Subsets

### CAR + MSMARCO Collection
Together CAR and MSMARCO have a size of 6,13G, so downloading will take a while. You can use the collection as followed:
```python
collection = load_dataset('trec-cast-2019-multi-turn', 'test_collection')
```

The collection has the following data format:
```
docno: str
The document id format is [collection_id_paragraph_id] with collection id and paragraph id separated by an underscore.
The collection ids are in the set: {MARCO, CAR}. E.g.: CAR_6869dee46ab12f0f7060874f7fc7b1c57d53144a
text: str
The content of the passage.
```

#### Sample
Instead of using the entire data set, you can also download a sample set containing only 200,000 items:
```python
collection = load_dataset('trec-cast-2019-multi-turn', 'test_collection_sample')
```

### Topics
You can get the topics as followed:
```python
topics = load_dataset('trec-cast-2019-multi-turn', 'topics')
```

The topics have the following dataformat:
```
qid: str
Query ID of the format "topicId_questionNumber"
history: str[]
A list of queries. It can be empty for the first question in a topic.
query: str
The query
```

### Qrels
You can get the qrels as followed:
```python
qrels = load_dataset('trec-cast-2019-multi-turn', 'qrels')
```

The qrels have the following data format:
```
qid: str
Query ID of the format "topicId_questionNumber"
qrels: List[dict]
A list of dictionaries with the keys 'docno' and 'relevance'. Relevance is an integer in the range [0, 4]
```

原始信息汇总

TREC Cast 2019 数据集概述

基本信息

• 语言: 英语 (en-US)
• 多语言性: 单语种
• 大小: 10M<n<100M
• 任务类别: 文本检索
• 任务ID: 文档检索

数据集统计

• 段落数量: 38,426,252
• 主题数量: 20
• 查询数量: 173

数据集子集

主要集合

• 名称: CAR + MSMARCO Collection

• 大小: 6.13GB

• 加载方式: python collection = load_dataset(trec-cast-2019-multi-turn, test_collection)

• 数据格式:

  • docno: 文档ID，格式为[collection_id_paragraph_id]
  • text: 段落内容

样本集合

• 大小: 200,000项
• 加载方式: python collection = load_dataset(trec-cast-2019-multi-turn, test_collection_sample)

主题数据

• 加载方式: python topics = load_dataset(trec-cast-2019-multi-turn, topics)

• 数据格式:

  • qid: 查询ID，格式为"topicId_questionNumber"
  • history: 查询历史列表
  • query: 查询内容

Qrels数据

• 加载方式: python qrels = load_dataset(trec-cast-2019-multi-turn, qrels)

• 数据格式:

  • qid: 查询ID，格式为"topicId_questionNumber"
  • qrels: 包含docno和relevance键的词典列表，relevance为[0, 4]范围内的整数