SnipsNSD, ATIS-NSD

本研究构建了两个公共数据集SnipsNSD和ATIS-NSD，用于任务导向对话系统中的新颖槽位检测（NSD）任务。这些数据集基于原有的槽位填充数据集Snips和ATIS，旨在通过识别未知或超出域的槽位类型来增强对话系统的能力。数据集的创建涉及三种策略：替换、掩码和移除，以模拟实际应用场景中逐渐出现的潜在新颖槽位。这些数据集的应用领域包括避免执行错误操作和发现未来对话系统开发的新实体类型，从而提高系统的功能和技能。

This study constructs two public datasets, SnipsNSD and ATIS-NSD, for the novel slot detection (NSD) task in task-oriented dialogue systems. Built upon the original slot filling datasets Snips and ATIS, these datasets aim to enhance the capability of dialogue systems by identifying unknown or out-of-domain slot types. Three strategies, namely replacement, masking, and removal, are employed in the creation of these datasets to simulate the gradually emerging potential novel slots in real-world application scenarios. The application scenarios of these datasets include avoiding erroneous operations and discovering new entity types for future dialogue system development, thereby improving the system's functionalities and skills.