CRAG Dataset Documentation

Dataset Version Information

DATASET_DESCRIPTION_VERSION: v3
DATASET_VERSION: v3

Overview

The CRAG dataset is designed to support the development and evaluation of Retrieval-Augmented Generation (RAG) models. It consists of two main types of data:

Question Answering Pairs: Pairs of questions and their corresponding answers.
Retrieval Contents: Contents for information retrieval to support answer generation.

Retrieval contents are divided into two types to simulate practical scenarios for RAG:

Web Search Results: For each question, up to 50 full HTML pages are stored, retrieved using the question text as a search query. For Task 1, 5 pages are randomly selected from the top-10 pages. These pages are likely relevant to the question, but relevance is not guaranteed.
Mock KGs and APIs: The Mock API is designed to mimic real-world Knowledge Graphs (KGs) or API searches. Given some input parameters, they output relevant results, which may or may not be helpful in answering the user's question.

Download CRAG Data

Task #1: Retreival Summarization Task Page
Task #2: Mock API Repository
Task #3: End to End Retreival Augmentation Task Page

Data Schema

Field Name	Type	Description
`interaction_id`	string	A unique identifier for each example.
`query_time`	string	Date and time when the query and the web search occurred.
`domain`	string	Domain label for the query. Possible values: "finance", "music", "movie", "sports", "open". "Open" includes any factual queries not among the previous four domains.
`question_type`	string	Type label about the query. Possible values include: "simple", "simple_w_condition", "comparison", "aggregation", "set", "false_premise", "post-processing", "multi-hop".
`static_or_dynamic`	string	Indicates whether the answer to a question changes and the expected rate of change. Possible values: "static", "slow-changing", "fast-changing", and "real-time".
`query`	string	The question for RAG to answer.
`answer`	string	The gold standard answer to the question.
`alt_ans`	list	Other valid gold standard answers to the question.
`split`	integer	Data split indicator, where 0 is for validation and 1 is for the public test.
`search_results`	list of JSON	Contains up to `k` HTML pages for each query (`k=5` for Task #1 and `k=50` for Task #3), including page name, URL, snippet, full HTML, and last modified time.

Search Results Detail

Key	Type	Description
`page_name`	string	The name of the webpage.
`page_url`	string	The URL of the webpage.
`page_snippet`	string	A short paragraph describing the major content of the page.
`page_result`	string	The full HTML of the webpage.
`page_last_modified`	string	The time when the page was last modified.