ONTOURL
收藏资源简介:
ONTOURL是一个全面的基准数据集,旨在评估大型语言模型在处理本体知识方面的能力。该数据集由来自8个领域的40个本体中生成的58,981个问题组成,涵盖了15个任务,旨在评估模型在理解、推理和学习三个维度的能力。这些任务包括概念理解、结构知识、逻辑推理、结构构建和概念对齐等方面。ONTOURL的数据来源包括科学、健康医疗、商业金融、地球环境、艺术娱乐、食品农业、人类社会和法律等领域。数据集的创建过程涉及从本体中提取元素,形成问题,添加难度,以及控制数据质量等步骤。该数据集为评估LLM在处理结构化符号知识方面的能力提供了一个重要的基准。
ONTOURL is a comprehensive benchmark dataset developed to evaluate the capabilities of large language models (LLMs) in handling ontological knowledge. It comprises 58,981 questions generated from 40 ontologies across 8 distinct domains, and covers 15 tasks designed to assess model performance across three core dimensions: comprehension, reasoning, and learning. These tasks cover aspects such as concept comprehension, structural knowledge, logical reasoning, structure construction, and concept alignment. The 8 source domains of the ONTOURL dataset include science, healthcare, business and finance, earth and environment, art and entertainment, food and agriculture, human society, and law. The dataset creation process involves steps such as extracting elements from ontologies, formulating questions, adding difficulty levels, and controlling data quality. This dataset serves as a critical benchmark for evaluating LLMs' capabilities in processing structured symbolic knowledge.




