Representing visual aesthetic tastes for web pages in computable form

Research suggests that the visual attractiveness of a website is a strong determinant of users’ online behavior. However, designing visually attractive websites is difficult. Among other challenges, web designers have to balance aesthetic design decisions against usability objectives, and account for the increasingly diverse range of display platforms (smartphones, tablets, etc.) on which web pages can now be viewed on. Intelligent design support systems are a potential solution to this problem. However, to intelligently assist designers in creating attractive web pages, computer systems need to possess computable representations of human visual aesthetic tastes. Using machine-extracted style-based features and visual aesthetic judgements from 18 volunteers on 56 websites, we present preliminary results that suggest that such representation may be possible. INTRODUCTION AND MOTIVATION Recent research suggests that the visual aesthetics of web interfaces is a strong determinant of users’ online behavior. In a study assessing trustworthiness of online health information sites, visually unappealing sites were rejected within a few seconds, whereas more appealing sites were scrutinized for content before being either accepted or rejected [12]. In the context of e-commerce, research has also indicated that the visual aesthetic quality of websites impacts intention of consumers to purchase products [1, 2]. The cited studies suggest that visual design choices such as low color contrast Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. between text and background, large amount of text on a page and poor use of color are some sources for variation in user behavior. Designing aesthetically pleasing web sites is challenging for several reasons. Cultural differences in aesthetic tastes pose a significant—possibly insurmountable—challenge to designers aiming to appeal to broad audiences [9]. Furthermore, designers must also balance aesthetic design decisions against other design objectives. For example, designers must make decisions about the amount of text on a page versus the (popular) aesthetic preference for “clean designs”, or decisions about the load time of a web page against the quality of graphics to include on the page. The diversity of computing platforms has also increased the need for websites that “look good” not only on a laptop screen but also on a 3.5" smartphone screen and a 10" tablet screen, making the job of a web designer even more challenging. A potential solution to these challenges are intelligent design support tools that can assist in creating visually attractive designs. These tools should be able to provide designers with creative inspiration through presentation of appropriate examples [10], directly suggest aesthetics-based improvements such as “reduce the amount of text,” “increase the color contrast between background and text,” “reduce the number of images” etc., to automatically generating more visually attractive alternatives of the current design. However, a prerequisite for such systems is the capacity to represent visual aesthetic tastes in a computable form. More specifically, in order to offer the envisioned level of design support, computer systems should be able to represent human visual aesthetic tastes in terms of variables in the design space. Using visual aesthetic judgements from 18 volunteers on 56 web designs, along with statistical learning techniques, we demonstrate that such representation may be feasilbe. In the remainder of this paper we summarize prior work in the area of statistical learning of “visual appeal” for web designs. We then describe an experiment to investigate the possibility of representing visual aesthetic tastes for web designs in a computable form. We conclude with a discussion of our findings and recommendations for future work in this area.