Companies such as Netflix and Amazon use recommendation engines to suggest products or content based on user behavior. For example, when you purchase an item on Amazon, you are shown other products that may interest you. Similarly, Netflix suggests TV shows and movies based on what you have watched.
We have applied the same concept to premium technology adoption across the internet. Our system predicts which technologies a website may consider using by analyzing the behavior of other websites with similar premium technology profiles.
How it Works
- We identify websites with significant investment in technology and determine which technologies they use.
- We measure the overlap between these technologies to establish the level of similarity between websites.
- The greater the overlap, the stronger the recommendation for premium technologies that one site is using and the other is not.
Why We Don’t Use a Traditional AI Model
You might wonder why we do not use a machine learning approach such as Matrix Factorization for our recommendation system. We tested several AI models, but the results were poor. These models also rely on complex internal parameters, often referred to as “magic numbers,” that are difficult to interpret or control.
For example, the MatrixFactorizationTrainer documentation includes the phrase “For better results use the following,” followed by two unexplained variables. There is no transparency about why these values improve results, which makes the model difficult to validate and trust.
Our chosen algorithm, by contrast, follows a clear and well-understood methodology. We know exactly how it works, how it is implemented, and why it produces meaningful outcomes.
Does It Actually Work?
Yes. For instance, the model identifies that websites likely to adopt Magento Enterprise often already use Magento. This relationship was not manually defined or trained into the system—it was discovered through the data itself.
Recommendations API
The recommendations can be accessed programmatically through our API, which provides both JSON and XML responses.
https://api.builtwith.com/recommendations-apiBuild Your Own Models
You can create your own models using BuiltWith Datasets, which provide the underlying data required for training and experimentation.