If you ever purchased an item online, you’re already intimately aware of recommendation systems. Perhaps a recommended product followed you while browsing webshops and online marketplaces. Or perhaps you searched for a screwdriver and, as a result, you see different hardware brands and drills everywhere. This, of course, is because artificial intelligence creates more useful recommendations with large data sets.

At this year’s Internet Hungary conference, Gábor Gönczy – the CEO of Stylers Group – discussed the operating principle behind AI’s recommendation systems. Below, we outline his main talking points:

First, to understand how recommendation systems operate daily, it is worthwhile to take a look at the different types of recommendation systems:

Content-based recommendation systems are based on previous evaluations of users. However, in many cases, this is not sufficient enough because [every user does not write]reviews. Some people are just excited about their new device and – after they start to use it – they never look back. Context-based recommendation systems go one step up and take into account the environmental variables which may affect user behaviour (for example if they’re married, moved to a new home, or just had a baby, etc.). Collaborative systems go even further. Their recommendations are based on even similar users’ previous behaviour, and they focus on the user (user-based) or the searched item (item-based).”

Hybrid recommendation systems combine content-based, context-based, and collaborative recommendation systems  for truly useful recommendations. This solution is the most beneficial for both the customer and seller. Gábor demonstrated the functioning of these systems by comparing a static and a collaborative AI recommendation system:

”Static systems recommend [an item] from a category that you’ve already purchased from, whereas collaborative AI systems also take into account other similarities.” – said Gábor.

A true collaborative AI system takes into account thousands of identifying factors, which give it nearly mind-reading capabilities, as it uses this data to predict our needs. Moreover, it works independently – we don’t have to add connections to the code manually, the system learns them by itself. The more the information, the better!

Obviously, these systems can only work with the data they receive from us. This fact divides internet users into two groups: people who hide their privacy at all costs and the less cautious people who press “OK” everywhere without a second thought.

These two are extremes, certainly, but it’s clear that recommendation systems have a more difficult job with the first group. Nevertheless, thanks to the information they are obligated to share, collaborative AI solutions still have enough data to see some of their needs. Consequently, these systems are more efficient for them too. And, in the case of the second group, the system will know when it’s time to switch from hiking tips to diaper advertisements – before you even realize that you’re going to have a baby.