xNot360
收藏资源简介:
xNot360数据集是由日本国立情报学研究所创建,专门用于评估GPT模型在自然语言中否定检测的能力。该数据集包含360个样本,涉及180个正样本和180个负样本,每个样本由5至20个单词组成,旨在通过多样化的句子结构全面测试模型处理否定的情况。数据集的创建旨在解决现有模型在处理否定时的局限性,特别是在高风险领域如医疗、法律和科学中的应用。通过这一数据集,研究者能够更深入地理解模型在处理复杂逻辑任务如否定检测时的表现,从而推动自然语言理解技术的进步。
The xNot360 dataset was developed by the National Institute of Informatics of Japan, specifically designed to evaluate the negation detection capability of GPT models in natural language. This dataset includes 360 samples, with 180 positive samples and 180 negative samples. Each sample is composed of 5 to 20 words, aiming to comprehensively test the model's performance in handling negation through diverse sentence structures. The creation of this dataset targets addressing the limitations of existing models in negation processing, especially for their applications in high-risk domains such as healthcare, law, and scientific research. Using this dataset, researchers can gain a deeper understanding of the model's performance when dealing with complex logical tasks like negation detection, thus advancing the progress of natural language understanding technologies.

- 1A negation detection assessment of GPTs: analysis with the xNot360 dataset日本国立情报学研究所 · 2023年



