IFS-INTENT, IFS-RELATION

本研究发布了两个基准数据集：IFS-INTENT和IFS-RELATION，用于增量少样本文本分类任务。IFS-INTENT源自BANKING77，包含77个意图类别，模拟银行客户服务场景。IFS-RELATION则从FewRel转换而来，涉及多领域关系分类。这两个数据集均未提供开发集，以模拟真实世界中缺乏额外标注数据的情况。数据集的创建旨在解决低资源环境下，系统如何通过少量标注样本来持续学习和识别新类别的问题。

This study presents two benchmark datasets, IFS-INTENT and IFS-RELATION, for incremental few-shot text classification tasks. IFS-INTENT is derived from BANKING77, which contains 77 intent categories and simulates the banking customer service scenario. IFS-RELATION, converted from FewRel, covers multi-domain relation classification. Neither of the two datasets provides a development set, to simulate the real-world scenario where additional labeled data is scarce. The creation of these datasets aims to address the problem of how systems can continuously learn and recognize new categories with only a small number of labeled samples in low-resource environments.