Frames

Frames数据集是由加拿大Maluuba Research Montreal创建的，包含1369个人与人的对话，平均每个对话有15轮。该数据集旨在研究目标导向对话系统中记忆的作用。数据集内容涉及在Wizard-of-Oz设置下收集的对话，用户与扮演对话系统角色的‘巫师’进行交流，寻找满足特定约束如目的地和预算的度假套餐。数据集的创建过程涉及20天的数据收集，共有12名参与者。Frames数据集主要应用于研究对话管理中的记忆机制和自然语言生成中的信息呈现，旨在解决对话系统中多状态同时跟踪的挑战。

The Frames dataset was developed by Maluuba Research Montreal, Canada. It consists of 1,369 human-to-human dialogues, with an average of 15 conversational turns per dialogue. This dataset is intended to investigate the role of memory in goal-oriented dialogue systems. The dialogues were collected under the Wizard-of-Oz paradigm, in which users interact with a "wizard" portraying a dialogue system to seek vacation packages that satisfy specific constraints including destination and budget. The data collection process spanned 20 days and involved a total of 12 participants. The Frames dataset is primarily applied to research on memory mechanisms in dialogue management and information presentation in natural language generation, aiming to address the challenge of simultaneous tracking of multiple states in dialogue systems.