Correcting Visual Design Errors in Presentation Slides

Abstract

When creating presentation slides, users often face difficulties in staying consistent with a presentation’s visual style and may be required to refer to previous slides. This may slow the user’s work or may lead to visual design errors when users create elements inconsistent with a presentation’s visual style. In our project, a machine learning algorithm – specifically decision trees – is used to predict attributes of visual elements, including font size and the positioning of elements in relation to neighboring elements. The user can use these predictions to adjust the attributes of elements to match the presentation’s visual style, potentially helping them more quickly complete tasks and identify potential errors. We implemented this approach as a Google Slides Add-on. In the future, we hope to enable additional human-computer collaboration where users may edit the visual styles learned by the system to reflect their intent.