User behavior discovery from low‐level software execution log

Abstract

During the execution of software, its execution log can be recorded. This provides valuable information on software runtime analysis, based on which software developers and vendors can find useful insights into how users behave and enable software usability improvements. However, the recorded software execution log usually is in such a fine granularity (usually in the method‐call level) that it cannot reflect real user behavior explicitly. To this end, we propose a supervised learning approach to detect high‐level user operation behavior from low‐level software execution log. More specifically, we first construct a behavioral pattern (represented by Petri nets) for each user operation. Then, we apply an alignment‐based matching approach which takes a set of behavior patterns and a software execution log as input to generate an abstract user operation log. Next, existing process discovery approaches are used to discover user behavior model from the obtained user operation log. The proposed approaches are supported by tools implemented in the open‐source process mining toolkit ProM. By experimental analysis, we demonstrate that the proposed approach facilitates the discovery of insightful user behavior with high accuracy from the software execution log. © 2018 Institute of Electrical Engineers of Japan. Published by John Wiley & Sons, Inc.