Machine Learning Based Reports

Netflix, Amazon and many other companies use 'recommendation engines'. You may have seen this when you buy something on Amazon it generally recommends other products to you. If you have watched Netflix it will also offer recommended TV shows and movies.

We've applied the same process to premium technology usage across the entire internet. This lets us predict what technologies a website might consider using based on what websites with similar websites (by premium technology usage) are doing.

How it Works

• For every website with a substantial technology spend we find what technologies match
• We use the amount of cross over to determine the weight that website has for its recommendations. So if a website matches on a lot of technologies, any premium technologies the site is not using are more strongly recommended than if the websites match on fewer premium technologies

Why don't we use a true AI model like Matrix Factorization to do this?
We tried but the dataset results was very poor and the AI models use 'magic numbers' that we have no control over or any real understanding of what they do. A good example of this is in the MatrixFactorizationTrainer code here there's a reference "For better results use the following" and then two variables without any explanation of why they produce better results.

The algorithm we use has a very clear methodology, we understand how it's implemented and why it works.

Does it truly actually work?
Yes we think so - for example the 'future customers' for Magento Enterprise generally already use 'Magento' as a technology. We did not tell the system that these two technologies are related or that Magento Enterprise customers would logically start with Magento.

Recommendations API Endpoint
Yes you can access the data via an API endpoint providing JSON and XML responses - https://api.builtwith.com/recommendations-api.

Create your own models
You need to use BuiltWith Datasets to get the underlying data to build your own models.