coastalcph/medical-bios

该数据集包含带有职业和二元性别标签的英文传记，主要用于职业分类任务，并可用于研究性别偏见。数据集包括10,000个传记，分为训练集、开发集和测试集，目标职业为5种医疗职业（心理学家、外科医生、护士、牙医、医生），源自De-Arteaga等人（2019）的研究。此外，数据集还包含100个传记的人类理性注释，分为非对比和对比两种设置。在非对比设置中，注释者被要求找到传记中描述某人为何从事某一职业的理由；在对比设置中，注释者被要求找到传记中描述某人为何从事某一职业而非另一职业的理由。This dataset contains English biographies annotated with occupation and binary gender labels, primarily intended for occupation classification tasks and gender bias research. The dataset includes 10,000 biographies split into training, development, and test sets, with five target medical occupations: psychologist, surgeon, nurse, dentist, and physician, derived from the study by De-Arteaga et al. (2019). Additionally, it provides human rational annotations for 100 biographies, divided into two settings: non-contrastive and contrastive. In the non-contrastive setting, annotators are asked to identify the reasons in the biography explaining why the subject pursued a particular occupation. In the contrastive setting, annotators are required to find the reasons in the biography explaining why the subject chose a particular occupation rather than another.

原始信息汇总

数据集描述

该数据集包含英文传记，标注了职业和二元性别。这是一个职业分类任务，可以研究性别偏见。数据集包含10,000条传记（8k训练集/1k开发集/1k测试集），针对5种医疗职业（心理学家、外科医生、护士、牙医、医生），源自De-Arteaga et al. (2019)。我们收集并发布了100条传记的人类理由注释，分为非对比和对比两种设置。在非对比设置中，注释者被要求找到理由：“为什么以下简短传记中的人被描述为L？”，其中L是黄金标签职业，例如护士。在对比设置中，问题是“为什么以下简短传记中的人被描述为L而不是F”，其中F（对照）是另一种医疗职业，例如医生。

数据集结构

我们提供standard版本的数据集，示例如下：

json { "text": "He has been a practicing Dentist for 20 years. He has done BDS. He is currently associated with Sree Sai Dental Clinic in Sowkhya Ayurveda Speciality Clinic, Chennai. ... ", "label": 3, }

以及新策划的包含人类理由的子集，称为rationales，示例如下：

json { "text": "She is currently practising at Dr Ravindra Ratolikar Dental Clinic in Narayanguda, Hyderabad.", "label": 3, "foil": 2, "words": [She, is, currently, practising, at, Dr, Ravindra, Ratolikar, Dental, Clinic, in, Narayanguda, ,, Hyderabad, .] "rationales": [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], "contrastive_rationales": [0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0] "annotations": [[0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], ...] "contrastive_annotations": [[0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], ...] }

使用方法

加载standard版本的数据集：

python from datasets import load_dataset dataset = load_dataset("coastalcph/medical-bios", "standard")

加载包含人类理由的新策划子集：

python from datasets import load_dataset dataset = load_dataset("coastalcph/medical-bios", "rationales")

引用

@inproceedings{eberle-etal-2023-rather, title = "Rather a Nurse than a Physician - Contrastive Explanations under Investigation", author = "Eberle, Oliver and Chalkidis, Ilias and Cabello, Laura and Brandl, Stephanie", booktitle = "Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing", year = "2023", address = "Singapore", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2023.emnlp-main.427", }