OPPO小布对话文本语义匹配数据集

通过对闲聊、智能客服、影音娱乐、信息查询等多领域真实用户交互语料进行用户信息脱敏、相似度筛选处理得到，数据集主要特点是文本较短、非常口语化、存在文本高度相似而语义不同的难例。该数据集所有标签都有经过人工精标确认。

This dataset is constructed from real-world user interaction corpora across multiple domains including chitchat, intelligent customer service, audio-visual entertainment, information query and more, after processing steps of user information de-identification and similarity screening. It is characterized by short text length, highly colloquial language, and challenging cases where texts are highly similar yet have distinct semantics. All labels within the dataset have been manually refined and verified.