Data Compression Algorithms in Analysis of UI Layouts Visual Complexity

• Published in: Perspectives of System Informatics Vol. 11964; pp. 167 - 184
• Main Authors: Bakaev, Maxim, Goltsova, Ekaterina, Khvorostov, Vladimir, Razumnikova, Olga
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Algorithmic complexity; Human-Computer interaction; Information processing; Static UI analysis
• ISBN: 9783030374860; 3030374866
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-37487-7_14

Measuring visual complexity (VC) of human-computer user interfaces (UIs) sees increasing development, as VC has been found to affect users’ cognitive load, aesthetical impressions and overall performance. Spatial allocation and ordering of UI elements is the major feature manipulated by an interface designer, and in our paper we focus on perceived complexity of layouts. Algorithmic Information Theory has justified the use of data compression algorithms for generating metrics of VC as lengths of coded representations, so we consider two established algorithms: RLE and Deflate. First, we propose the method for obtaining coded representations of UI layouts based on decreasing of visual fidelity that roughly corresponds to the “squint test” widely used in practical usability engineering. To confirm applicability of the method and the predictive power of the compression algorithms, we ran two experimental surveys with over 4700 layout configurations, 21 real websites, and 149 participants overall. We found that the compression algorithms’ metrics were significant in VC models, but the classical purely informational Hick’s law metric was even more influential. Unexpectedly, algorithms with higher compression ratios that presumably come closer to the “real” Kolmogorov complexity did not explain layouts’ VC perception better. The proposed novel UI coding method and the analysis of the compression algorithms’ metrics can contribute to user behavior modeling in HCI and static testing of software UIs.