Assessing reproducibility in screenshot-based task mining: A decision discovery perspective

Abstract

This reproducible paper serves as a companion paper to prior work introducing a novel Task Mining framework that incorporates UI screenshots as supplementary data to enhance the interpretability of human decision-making in Robotic Process Automation (RPA). This framework enriches the traditional User Interface (UI) log — typically composed of timestamped events such as mouse clicks and keystrokes — with image data, allowing for a more detailed process model to be discovered, particularly in the context of decision-making rules. The aim of this reproducibility paper is to provide a detailed, step-by-step reproducibility protocol to replicate the Task Mining framework’s core methodology, including data processing, extraction of features within screenshots, and the construction of decision trees based on enriched UI logs to reproduce the results obtained on the original paper on a set of experiments designed to validate the accuracy and efficacy of the framework across varying UI log sizes and interface complexities. Finally, we make an argument that the results reported in our primary work can be considered weakly reproducible.