Learning Users Habits: the Ape Project

This paper proposes new results in the eld of software assistants helping users of interactive tools in the task of automatically performing repetitive tasks. We propose an innovative integration of such an assistant into an interactive programming environment. In this context, learning to recognize situations in which repetitive tasks occur is diicult because languages describing users actions are complex and because fast learning is mandatory. To achieve this goal we propose an agent-based software assistant that includes a new learning algorithm. The algorithm is incremental, has a very low training time and is able to handle diierent description languages. In the context of that application, it provides better results than related machine learning algorithms, making our assistant actually usable and eecient. Our software assistant is embodied into the ParcPlace VisualWorks (Smalltalk) programming environment and accessible through the net. Its architecture and algorithm could be adapted with beneet to any application involving human-computer interaction.