Verbal-ConvQuestions

Verbal-ConvQuestions是由波恩大学等机构创建的数据集，专注于知识图谱上的对话式问答系统，特别是在多轮对话中提供口头化的答案。该数据集包含11200条对话，每条对话涉及复杂的查询，如聚合、组合性、时间推理和比较。创建过程中，研究人员采用了半自动框架，利用回译技术生成多样的口头化答案。此数据集的应用领域主要集中在提升语音助手（如Siri、Alexa和Google Assistant）的用户交互体验，解决现有系统在自然语言交互中的不足。Verbal-ConvQuestions is a dataset developed by institutions including the University of Bonn, focusing on conversational question answering systems over knowledge graphs, with a particular emphasis on generating verbalized answers in multi-turn dialogues. This dataset includes 11,200 dialogues, each revolving around complex queries such as aggregation, compositional reasoning, temporal reasoning, and comparison. During its development, researchers adopted a semi-automatic framework that leveraged back-translation technology to generate diverse verbalized answers. The main application scenarios of this dataset aim to enhance user interaction experiences of mainstream voice assistants like Siri, Alexa, and Google Assistant, and address the shortcomings of existing systems in natural language interactions.