PhotoBook数据集

PhotoBook数据集是由阿姆斯特丹大学的对话建模小组开发的一个大规模视觉基础对话数据集，旨在研究对话过程中共享对话历史的积累。该数据集包含2500个英语任务导向对话，涉及两个在线参与者通过书面聊天交换信息，以识别他们各自相册中的共享图片。数据集的创建过程涉及一个合作游戏，参与者需利用视觉上下文和先前建立的指称表达来指称图片。该数据集的应用领域主要集中在开发更一致和有效的对话模型，特别是在解决指称解析问题和理解对话中的共同基础方面。

The PhotoBook Dataset is a large-scale visually grounded dialogue dataset developed by the Dialogue Modeling Group at the University of Amsterdam, which aims to investigate the accumulation of shared conversation history during dialogue interactions. This dataset includes 2,500 English task-oriented dialogues, where two online participants exchange information via written chat to identify shared images present in their respective photo albums. The creation of the dataset is based on a cooperative game, requiring participants to leverage visual context and previously established referring expressions to refer to specific images. The primary application scope of this dataset focuses on developing more consistent and effective dialogue models, especially for addressing reference resolution issues and understanding common ground in conversational contexts.