BioNLI

BioNLI是一个专为生物医学领域设计的自然语言推理数据集，由石溪大学的研究团队创建。该数据集包含40,459条数据，主要来源于PubMed的摘要，涵盖了生物医学机制与实验证据的配对。数据集的创建过程采用了半监督方法，通过九种策略自动生成负例，确保了数据集的多样性和复杂性。BioNLI数据集的应用领域主要集中在生物医学文本的理解和推理，旨在帮助解决生物医学机制是否得到实验证据支持的问题。

BioNLI is a natural language inference dataset specifically designed for the biomedical domain, developed by a research team from Stony Brook University. This dataset contains 40,459 instances, primarily sourced from PubMed abstracts, and covers pairs of biomedical mechanisms and experimental evidence. The dataset was constructed using a semi-supervised method, where negative examples were automatically generated through nine strategies to ensure its diversity and complexity. The main application areas of the BioNLI dataset focus on biomedical text understanding and inference, aiming to assist in addressing the problem of whether a biomedical mechanism is supported by experimental evidence.