Understanding how developers work on change tasks using interaction history and eye gaze data

Abstract

Developers spend a majority of their efforts searching and navigating code with the retention and management of context being a considerable challenge to their productivity. We aim to explore the contextual patterns followed by software developers while working on change tasks such as bug fixes. So far, only a few studies have been undertaken towards their investigation and the development of methods to make software development more efficient. Recently, eye tracking has been used extensively to observe system usability and advertisement placements in applications and on the web, but not much research has been done on context management using this technology in software engineering and how developers work. In this thesis, we analyze an existing dataset of eye tracking and interaction history that were collected simultaneously in a previous study. We look into exploring navigational patterns of developers while they solve tasks. Our goal is to use this dataset to determine if we can perform prediction and recommendations solely based on eye gaze patterns. In order to do this, we conduct three experiments on Microsoft Azure on developer expertise recommendation and class recommendation for developers using only eye tracking data. Our results are quite promising. We find that eye tracking data can be used to predict expertise of developers with 85% accuracy. It is further able to recommend classes with good performance (a normalized discounted cumulative gain, NDCG ranging between 0.85 and 0.88). These findings are discussed with a view to designing systems that can adapt to the individual user in real time and make intelligent adaptive suggestions while developers work.