Extending the Gmail User-Interface to Leverage Prediction of Response Times and Hierarchical Recipients

Abstract

Recent work has shown how the Gmail user-interface can be extended to make use of algorithms for predicting a flat list of email recipients. We build on this work by addressing how this user-interface be extended to make use of external algorithms for predicting (a) a hierarchical list of email recipients, and (b) the expected time to get a response from a recipient. We have augmented the existing view for sending messages to allow users to view predictions of response times and hierarchical recipients, and use one click to select a subgroup of predicted recipients. We have also added new folder- specific commands to the views for browsing folders that allow users to use one click to highlight, un-highlight, and select messages whose responses have not arrived or been sent within the predicted times. We have also developed a new configuration view for determining which predictions should be displayed and how often new training data should be used to change the prediction model. The user-interface is implemented as a Chrome extension that communicates with an external server to receive predictions and send training data. This architecture decouples the implementation of the user-interface and algorithms. However, our extension is intimately tied to the extended user-interface, as it reads and augments existing views and menus of Gmail.