Designing websites with Deep Learning and Triplet Loss
Web layout optimization is one of the main problems for website creators, designers, and automated cloud services. There are many applications that one can think of.
Let’s say you already designed a website but now you just want to have it on a mobile or tablet, and there are so many sizes…One application involves taking a website, made by a user, and restructuring it for a mobile web browser.
In order to do so, we designed an optimization process that learns how to take a website and automatically, without any human involvement, generate the website layout for the desired device (like mobile) or vice versa.
As part of the optimization process, it is required to identify groups of web components, which should be processed sequentially when placing them on the mobile page.
As with every AI model, we generated a dedicated dataset to be able to build a model. Therefore we had to make a few assumptions:
- A page is processed top-down left-right. Starting with the top left most component on the page, the components are placed consecutively on the mobile page
- We have dedicated annotators for the task
- There are millions of websites available for us to learn with
Now for the most important part: The Data!
