IRT2 Benchmark
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/lavis-nlp/irt2
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是一个开放的基准测试，涵盖了不同规模的知识图谱，包含了低质量的文本提及，并支持三元组补全和排序的任务。此外，该数据集为评估提供了多个分割，旨在挑战模型在数据稀缺程度不同的环境下的表现。其规模多样，包括微小、小型、中型和大型，任务涵盖了诱导式链接预测和排名。
This dataset is an open benchmark encompassing knowledge graphs of various scales, which includes low-quality textual mentions and supports tasks including triplet completion and ranking. Additionally, it provides multiple evaluation splits designed to assess model performance in environments with varying degrees of data scarcity. The dataset offers variants across four scales: tiny, small, medium, and large, with tasks covering inductive link prediction and ranking.
