TriviaHG

TriviaHG是由因斯布鲁克大学创建的大型数据集，包含160,230个提示，对应16,645个来自TriviaQA数据集的问题。该数据集旨在支持自动提示生成，特别是在事实性问题回答领域。数据集内容涵盖人物、地点、实体、日期和数字等多种类型的问题，通过使用Copilot等大型语言模型生成提示。创建过程中，数据集利用了网络页面作为主要信息源，确保提示的准确性和相关性。TriviaHG的应用领域包括增强检索增强生成（RAG）解决方案和查询扩展方法，以及在娱乐领域的应用，如游戏节目Jeopardy。

TriviaHG is a large-scale dataset developed by the University of Innsbruck. It contains 160,230 prompts corresponding to 16,645 questions sourced from the TriviaQA dataset. This dataset is designed to support automatic prompt generation, particularly in the domain of factual question answering. Its contents cover various types of questions including persons, locations, entities, dates and numbers, with prompts generated using large language models such as Copilot. During the dataset creation process, web pages were utilized as the primary information source to ensure the accuracy and relevance of the prompts. Application domains of TriviaHG include enhancing retrieval-augmented generation (RAG) solutions and query expansion methods, as well as entertainment-related applications such as the game show Jeopardy.