Mobile app Tasks with Iterative Feedback (MoTIF)
收藏资源简介:
MoTIF是由波士顿大学创建的一个新型视觉语言导航数据集,专注于移动应用中的自然语言指令执行。该数据集包含6100条指令,涵盖了多种应用环境和复杂的操作序列,首次引入了指令可行性标注,包括二元可行性标签和细粒度的不可行原因标签。MoTIF旨在解决移动应用中指令执行的不确定性问题,通过迭代反馈机制,推动视觉语言导航技术在实际应用中的发展。数据集的应用领域广泛,包括移动应用设计、人机交互和文档理解等,为研究提供了丰富的资源和挑战。
MoTIF is a novel visual-language navigation dataset developed by Boston University, focusing on natural language instruction execution in mobile applications. This dataset includes 6,100 instructions covering diverse application scenarios and complex operation sequences, and introduces, for the first time, instruction feasibility annotations comprising binary feasibility labels and fine-grained infeasibility cause labels. MoTIF aims to address the uncertainty issue in instruction execution within mobile applications, and promotes the development of visual-language navigation technologies in real-world applications via an iterative feedback mechanism. The dataset has broad application domains including mobile application design, human-computer interaction and document understanding, providing abundant resources and research challenges for relevant studies.




