NLU-Evaluation-Data

本数据集名为NLU-Evaluation-Data，由赫瑞瓦特大学创建，包含25716条多领域（21个领域）的用户语音数据，涵盖64种意图和54种实体类型。数据集通过亚马逊Mechanical Turk收集，涉及多种场景，如设置闹钟、播放音乐等。创建过程包括数据收集、意图和实体类型标注，以及多轮验证以确保数据质量。该数据集主要用于评估和比较不同自然语言理解服务在构建对话代理中的性能，旨在解决如何选择和优化NLU服务的问题。

This dataset, named NLU-Evaluation-Data, was created by Heriot-Watt University. It contains 25,716 instances of user speech data spanning 21 distinct domains, encompassing 64 intents and 54 entity types. The dataset was collected through Amazon Mechanical Turk, covering diverse scenarios including setting alarms, playing music, and other daily use cases. Its development pipeline includes data collection, annotation of intents and entity types, and multi-stage validation to guarantee data quality. This dataset is primarily designed to evaluate and compare the performance of various natural language understanding (NLU) services when constructing conversational agents, with the goal of addressing the challenge of selecting and optimizing NLU services.