entity-switched-ner

本研究创建了名为‘entity-switched-ner’的数据集，由宾夕法尼亚大学等机构开发，旨在通过替换文本中的实体名称来测试命名实体识别（NER）模型的域内鲁棒性。数据集通过程序化方法扩展现有数据，包含多样化的实体及其在不同上下文中的表现。该数据集特别关注实体的国家来源差异，以评估模型在识别不同国家实体时的性能。此数据集的应用领域主要集中在提高NER系统的鲁棒性和公平性，解决模型在处理不同文化背景实体时的性能差异问题。

This study constructs the dataset named "entity-switched-ner", developed by institutions such as the University of Pennsylvania. This dataset is designed to test the in-domain robustness of Named Entity Recognition (NER) models by replacing entity names in textual data. It expands existing corpora through automated procedural methods, covering diverse entities and their occurrences across various contextual settings. This dataset specifically focuses on the national origin disparities of entities to evaluate the performance of models when recognizing entities from different countries. The primary application scenarios of this dataset are aimed at improving the robustness and fairness of NER systems, and addressing the performance discrepancies of models when processing entities from diverse cultural backgrounds.