The Postdiction as an approch to recommand commands based on interface state evolution

Users use only a limited subset of software commands. This article introduces a a novel approach to discovering commands in graphical user interfaces. Unlike previous approaches to recommend a command, which relies on command prediction - determining what the user wants to do in the future - our approach relies on command postdiction - explaining how a task in the past could have been performed more efficiently. Given the current state of the interface and previous states, our approach infers the command that would have been most appropriate and suggests it to the users. We discuss three case studies: shape, image and text editors. For each of them, we study the feasibility of the approach through proofs of concept. Finally, we discuss the challenges posed by deplying post-diction in existing software.