Discovering Automatable Routines from UI Logs via Sequential Pattern Mining

Abstract

Robotic Process Automation (RPA) is a rapidly evolving technology that allows us to auto-mate non-value adding tasks (i.e., routine tasks), such as transferring data from one appli-cation to another. The automation of such tasks allows us to reduce the number of errors that occur during its execution and decrease task execution time. However, RPA intended to be used for automation routines, but not for their discovering. This thesis proposes a method for discovering routines from user interaction log by exploiting sequential pattern mining techniques for dealing with noise within the log. Since it is essential to understand which of the discovered routines are automatable, the thesis proposes a method for measur-ing the routine automatability index (RAI). The method is based on identifying if the routine actions are automatable by obtaining dependencies between them, more precisely, data transformations and functional dependencies. A comparative evaluation with the existing approach on synthetic and controlled-setting datasets shows that the proposed method can discover candidate routines, identify action dependencies, and measure RAI with acceptable execution time. The proposed approach has been implemented in Java, integrated with the SPMF pattern mining tool, Foofah data transformation tool, and Tane algorithm for finding functional dependencies.