This Algorithm Can Predict How Exciting a Website Is
Machine learning makes "excitingness" measurable.
This is what an image with a score of 95 looks like. Image: EyeQuant
How exciting is this website? I'm not asking your opinion—the answer is 80 out of 100.
That's according to new algorithms from a Berlin-based machine learning startup that rate the design of web pages on their "excitingness." Letting it loose on Motherboard's homepage a few days ago gave us a score of 80, ahead of most other tech news sites (too bad, guys!).
Fabian Stelzer, cofounder of EyeQuant, explained that the algorithm comes to its score by predicting how a group of a few hundred humans would rate the design of a website from 0 to 100—which would be a more conventional way of measuring something that seems so subjective. "Basically you get the same kind of results that you could usually only get when you run a study that might take days or weeks," he said. The computer version takes three to five seconds.
EyeQuant has previously offered tools that measure where viewers will focus their attention on a page (you might have seen "heat maps" that show this) and the visual clarity of a design. This new capability will be added to its web-based software and API, which customers—often ecommerce firms looking to best market their products—can run their designs through before setting them live.
The EyeQuant team has a background in "visual neuroscience," i.e. studying the human visual system, and Stelzer says this formed the basis of methods they used to "train" the machine. They first did a lot of research with humans to see how they rated excitingness, then applied machine learning methods to the data. "That enables us to essentially teach the software which patterns are responsible for driving excitingness," said Stelzer.
He explained they fed the process with around 200 features, and though it's hard to give any real rules of thumb about what makes a site exciting, they noted a few trends. Large-scale photographs usually help to add excitement, while tables tend to reduce it—"even though the software doesn't know what a table is, it doesn't have a concept of it, it's just purely the design of a table," said Stelzer.
But a high excitingness score doesn't necessarily make a good website. Another indicator the team measures is the visual clarity of the design (Motherboard does pretty well on that too with a score of 60). Stelzer says there's actually no correlation between excitingness and clarity—you can have a super-exciting page that's not very visually clean, and vice-versa.
And what scores you want depends on the function of your page. Google has a very low excitingness score (5) but its cleanness is pretty useful for a search engine. On the other end of the spectrum, Stelzer singled out gaming sites as being exciting but not very clean. Among sites that score as both cluttered and unexciting are text-heavy technology favourites Reddit and Hacker News.
Ultimately, the idea is simply to give designers a glimpse of how viewers will respond that's based on more than a hunch. "Particularly with excitingness, we don't say you should always optimise and try to get the highest score," said Stelzer. "It's rather a way to make something measurable that until now wasn't measurable."
- web design
- motherboard show
- machine learning