Box2Go: Collaborative Interactive Infobox Filling

Infoboxes can be useful to quickly learn about the contents of text collections, but manually creating them is error-prone and time-consuming, and existing automatic approaches require training data or resources like ontologies that are not available for every domain. Moreover, they lack techniques for adaptation to the user. We therefore propose a system to automatically fill user-defined attributes of infoboxes with the human-in-the-loop which provides this adaptation, and works without training data and domain-specific resources. Our approach generalizes simple user feedback to explore a joint embedding space and find the correct values for the attributes. These structured representations of the texts can be used for collaborative exploration of text collections on the web. We provide a prototypic implementation for such a collaborative web application and demonstrate its usage.