TEMPREASON
收藏资源简介:
TEMPREASON数据集是由阿里巴巴达摩院创建,旨在评估大型语言模型的时间推理能力。该数据集包含三种时间推理级别的问题,覆盖了从634年到2023年的时间范围,总计约52,800条数据。数据集通过提取Wikidata知识库中的时间相关事实构建,适用于闭卷、开卷和推理问答设置。TEMPREASON旨在解决时间敏感问答任务中的挑战,特别是模型在不同时间范围内的表现差异。
The TEMPREASON dataset was developed by Alibaba DAMO Academy to evaluate the temporal reasoning capabilities of large language models. It contains questions across three levels of temporal reasoning, covering a time span from 634 CE to 2023, with a total of approximately 52,800 data instances. Constructed by extracting temporally relevant facts from the Wikidata knowledge base, the dataset is suitable for closed-book, open-book, and reasoning-based question answering settings. TEMPREASON aims to address the challenges in temporal-sensitive question answering tasks, particularly the disparities in model performance across different time ranges.

- 1Towards Benchmarking and Improving the Temporal Reasoning Capability of Large Language Models达摩院 · 2023年



