miracl/hagrid
收藏资源简介:
--- license: apache-2.0 language: - en pretty_name: HAGRID size_categories: - 1K<n<10K --- # HAGRID: A Human-LLM Collaborative Dataset for Generative Information-seeking with Attribution HAGRID (**H**uman-in-the-loop **A**ttributable **G**enerative **R**etrieval for **I**nformation-seeking **D**ataset) is a dataset for generative information-seeking scenarios. It is constructed on top of MIRACL 🌍🙌🌏, an information retrieval dataset that consists of queries along with a set of manually labelled relevant passages (quotes). ## Dataset Structure To load the dataset: ```python import datasets hagrid = datasets.load_dataset("miracl/hagrid", split="train") print(hagrid[0]) ``` It would show: ```json { 'query': ..., 'query_id': ..., 'quotes': [{ # a list of quotes that are manually labeled as relevant to the query 'docid': ..., 'idx': ..., 'text': ... }, ...] 'answers': [{ 'answer': ..., # the complete answer generated by LLM 'attributable': 1/0/None, # 1: attributable; 0: unattributable; None: unlabeled 'informative': 1/0, # 1: informative; 0: uninformative 'sentences': [{ # answers split into sentences 'index': ..., 'attributable': 0/1/None, 'informative': 0/1/None, 'text': ..., }, ...] }, ...] } ```
HAGRID数据集概述
基本信息
- 许可证: Apache-2.0
- 语言: 英语
- 数据集名称: HAGRID
- 数据集大小: 1K<n<10K
数据集描述
HAGRID(Human-in-the-loop Attributable Generative Retrieval for Information-seeking Dataset)是一个用于生成式信息检索场景的数据集。该数据集构建在MIRACL之上,MIRACL是一个包含查询及其一组手动标记的相关段落(引用)的信息检索数据集。
数据集结构
数据集的结构如下:
- 查询(query)
- 查询ID(query_id)
- 引用列表(quotes):手动标记为与查询相关的引用列表,每个引用包含文档ID(docid)、索引(idx)和文本(text)。
- 答案列表(answers):由LLM生成的完整答案,每个答案包含:
- 答案文本(answer)
- 可归属性(attributable):1表示可归属,0表示不可归属,None表示未标记
- 信息性(informative):1表示信息性,0表示非信息性
- 句子列表(sentences):答案分割成的句子,每个句子包含索引(index)、可归属性(attributable)、信息性(informative)和文本(text)。



