An approach based on suffix array to discover routines in user interaction logs
Export citation
Abstract
Robotic Process Automation (RPA) is a fast-advancing technology that allows organizations to automate repetitive work by the use of software robots. The number of candidate processes for automation may be vast and, henceforth, it raises questions such as which processes have higher priority for automation? and which of them can be automated by means of User Interaction (UI) routines? Selecting routines amenable for automation requires distinguishing between noise and relevant events and defining the boundaries thereof. From a technical point of view, the discovery of routines starts with a log capturing the UI tasks performed by a human and partitions it into segments that are presumably candidate patterns that can be composed into routines. Existing techniques, however, fail when the UI log contains multiple intertwined routines, yet such situation occurs in real-world scenarios. This project tackles this problem leveraging techniques stemmed from the field of periodic sequence mining and, more specifically from stringology. The result is a collection of novel algorithms tailored for the problem at hand which shows good performance, resilience to noise and has been evaluated with artificial and human generated UI logs.