From shell logs to shell scripts

Abstract

Analysing the use of a Unix command shell is one of the classic applications in the domain of adaptive user interfaces and user modelling. Instead of trying to predict the next command from a history of commands, we automatically produce scripts that automate frequent tasks. For this we use an ILP association rule learner. We show how to speedup the learning task by dividing it into smaller tasks, and the need for a preprocessing phase to detect frequent subsequences in the data. We illustrate this with experiments with real world data.