LREBench
收藏资源简介:
LREBench是由浙江大学与AZFT联合实验室知识引擎创建的数据集,包含8个覆盖不同语言、领域和上下文的实体关系抽取(RE)数据集。这些数据集旨在评估低资源环境下关系抽取系统的性能,特别关注极端少样本实例和长尾分布问题。数据集创建过程中,采用了多种方法如提示调优、数据平衡和数据增强等,以提高模型在资源有限情况下的泛化能力。LREBench的应用领域主要集中在信息抽取,特别是解决低资源环境下的关系抽取问题,推动相关技术向实际工业场景的过渡。
LREBench is a dataset developed by the Knowledge Engine of the Joint Laboratory of Zhejiang University and AZFT. It consists of 8 entity relation extraction (RE) datasets covering diverse languages, domains and contexts. These datasets are designed to evaluate the performance of relation extraction systems in low-resource scenarios, with a particular focus on extremely few-shot instances and long-tail distribution problems. During the dataset construction process, multiple methods such as prompt tuning, data balancing and data augmentation were adopted to enhance the generalization ability of models in resource-constrained settings. The application scenarios of LREBench mainly focus on information extraction, particularly addressing relation extraction problems in low-resource environments, and promoting the transition of related technologies to real-world industrial applications.

- 1Towards Realistic Low-resource Relation Extraction: A Benchmark with Empirical Baseline Study浙江大学 & AZFT 联合实验室知识引擎 · 2023年



