User Interface Optimization using Genetic Programming with an Application to Landing Pages

The design of user interfaces (UIs), such as World Wide Web pages, usually consists in a human designer mapping one particular problem (e.g., the demands of a customer) to one particular solution (i.e., the UI). In this article, a technology based on Genetic Programming is proposed to automate critical parts of the design process. In this approach, designers are supposed to define basic content elements and ways to combine them, which are then automatically composed and tested with real users by a genetic algorithm in order to find optimized compositions. Such a strategy enables the exploration of large design state-spaces in a systematic manner, hence going beyond traditional A/B testing approaches. In relation to similar techniques also based on genetic algorithms, this system has the advantage of being more general, providing the basis of an overall programmatic UI design workflow, and of calculating the fitness of solutions incrementally. To illustrate and evaluate the approach, an experiment based on the optimization of landing pages is provided. The empirical result obtained, though preliminary, is statistically significant and corroborates the hypothesis that the technique works.