QuanTemp++
收藏资源简介:
QuanTemp++是一个由自然数值主张、开放领域语料库以及每个主张对应的相关证据组成的真实数据集。该数据集旨在通过模拟人类事实核查员的方法,收集相关证据,确保没有时间泄露。数据集包含大约15k个自然数值主张,165.7k条记录的开放领域语料库,以及每个主张的相关性标签和真实性标签。QuanTemp++通过模拟人类事实核查员的搜索过程,收集了高质量的证据,有助于更真实地评估和理解自动事实核查中的检索和验证瓶颈。
QuanTemp++ is a real-world dataset consisting of natural numerical claims, an open-domain corpus, and relevant evidence corresponding to each individual claim. This dataset aims to gather high-quality relevant evidence by simulating the workflow of human fact-checkers, while eliminating temporal leakage. It contains approximately 15k natural numerical claims, an open-domain corpus with 165.7k records, as well as relevance labels and veracity labels for each claim. QuanTemp++ collects evidence via simulating the search process of human fact-checkers, enabling more authentic evaluation and comprehension of retrieval and verification bottlenecks in automated fact-checking.

- 1通过Stockholm University,Sweden; Independent Researcher; Delft University of Technology,Netherlands · 2025年



