Do you know we now have an on-line convention about product design developing? SPRINT will cowl how designers and product house owners can keep forward of the curve in these unprecedented instances.

Over the previous few years, articles and blog posts have began to ask some model of the identical query: “Why are all websites starting to look the same?

These posts normally level out some frequent design components, from massive photographs with superimposed textual content, to hamburger menus, that are these three horizontal strains that, when clicked, reveal a listing of web page choices to select from.

My colleagues Bardia Doosti, David Crandall, Norman Su and I have been finding out the history of the web after we began to note these posts cropping up. Not one of the authors had achieved any type of empirical examine, although. It was extra of a hunch that they had.

We determined to research the declare to see if there have been any fact to the notion that web sites are beginning to look the identical and, if that’s the case, discover why this has been taking place. So we ran a series of data mining studies that scrutinized almost 200,000 photographs throughout 10,000 web sites.

How do you even measure similarity?

It’s just about inconceivable to check all the web; there are over a billion websites, with many instances as many webpages. Since there’s no record of all of them to select from, performing a random pattern of the web is off the desk. Even when it have been doable, most individuals solely see a tiny fraction of these web sites usually, so a random pattern might not even seize the web that most individuals expertise.

We ended up utilizing the web sites of the Russell 1000, the highest U.S. companies by market capitalization, which we hoped can be consultant of traits in mainstream, company net design. We additionally studied two different units of websites, one with Alexa’s 500 most trafficked websites, and one other with websites nominated for Webby Awards.

As a result of we have been within the visible components of those web sites, as knowledge, we used photographs of their net pages from the Internet Archive, which usually preserves web sites. And since we wished to assemble quantitative knowledge evaluating hundreds of thousands of web site pairs, we would have liked to automate the evaluation course of.

To do this, we needed to decide on a definition of “similarity” that we might measure mechanically. We investigated each particular attributes like shade and format, in addition to attributes realized mechanically from knowledge utilizing synthetic intelligence.

For the colour and format attributes, we measured what number of pixel-by-pixel edits we must make to rework the colour scheme or web page construction of 1 web site into one other. For the AI-generated attributes, we skilled a machine studying mannequin to categorise photographs primarily based on which web site they got here from and measure the attributes the mannequin realized. Our previous work signifies that this does a fairly good job at measuring stylistic similarity, however it’s very tough for people to grasp what attributes the mannequin centered on.

How has the web modified?

We discovered that throughout all three metrics – shade, format and AI-generated attributes – the common variations between web sites peaked between 2008 and 2010 after which decreased between 2010 and 2016. Format variations decreased essentially the most, declining over 30% in that timeframe.