
How AI decides what we like
Topics covered:
How does Netflix know what series you like to watch? Why does an online store suggest products that might interest you? This is thanks to recommendation systems that use artificial intelligence to predict your preferences. Discover the secrets of their operation and see how much they influence your daily choices.
The 1990s - The modest beginnings of recommendation systems
The first recommendation systems appeared in the 1990s. Initially, they were mainly based on collaborative filtering - analyzing the preferences of many users to find similarities. If someone liked the same movies as you, the system assumed you would also like another movie that person rated highly.
However, this approach had its limitations. It required a large number of users for the recommendations to be accurate and struggled with new products that did not yet have many ratings.

The development of recommendation systems
Over time, recommendation systems became more advanced. They started using not only ratings but also other user data such as:
- purchase history;
- viewed pages;
- time spent viewing a product;
- demographic data - age, gender, or place of residence.
This allowed for better personalization of recommendations. The system could, for example, notice that you often buy books by a particular author or that you are interested in natural cosmetics. Such information enabled it to select products more accurately just for you.
Recommendation systems today
Today's recommendation systems are powerful tools that use vast amounts of data and advanced artificial intelligence algorithms. They can analyze not only your behavior but also the content of products - such as descriptions, tags, or photos.
As a result, recommendations are becoming increasingly accurate, often surprising the recipients. The system might suggest a book in your favorite genre by an unknown author or a recipe for a cuisine you like with an unusual ingredient.
Recommendation systems continually learn and adapt to your preferences. The more you use them, the better they understand your taste and the more accurately they meet your needs.
Create your AI-based solution with us.
Why are recommendation systems so important today?
Today, we have access to an enormous amount of information, products, and services. Searching through this vast amount of data for something interesting can be time-consuming and frustrating. How can a good recommendation system help you?
The most important advantages of such systems are:
- time savings - you don't have to browse hundreds of products as the system suggests the best ones;
- discovering new things - recommendations help you learn about new brands and products;
- tailored offers - thanks to personalization, the system's suggestions are closer to your tastes and needs;
- convenience - recommendations often appear automatically, without the need to enter queries in the search engine.
Recommendation systems are invaluable in the age of information overload. Therefore, they are becoming increasingly popular among both users and website owners.
The basics of machine learning in recommendation systems
The heart of recommendation systems is machine learning algorithms. They enable the computer to draw conclusions from data about your preferences and find matching products. Let's take a closer look at some key concepts.
k-Nearest Neighbors algorithm (k-NN)
A popular algorithm used in recommendation systems is the k-nearest neighbors (k-NN) algorithm. It involves finding k users most similar to you in terms of preferences, your nearest neighbors (NN). Then, the AI recommends products they like that you haven't discovered yet.
Let's say you recently watched only documentaries on Netflix. The k-NN algorithm will find other users who also often choose this genre. If most of them like a series you haven't seen yet, the system will recommend it to you.
Fuzzy logic
Unlike classical logic, where something can only be true or false (0 or 1), fuzzy logic allows for intermediate states. How does this apply to recommendations on websites? Fuzzy logic determines how well a product matches your preferences.
Suppose you like thrillers but not necessarily horrors. For a system based on fuzzy logic, a thriller will closely match your preferences (e.g., 70%), while a horror will match less closely (e.g., 30%).
Image recognition
Modern recommendation systems can also analyze images, such as product photos. They use neural networks that "recognize" what is in the picture.
If you look at a few floral dresses in an online store, the system will analyze these images and understand that you are interested in that pattern. It will then be able to find other floral clothes and suggest them to you.
The possibilities of artificial intelligence are not limited to recognizing specific objects in pictures. It also understands more abstract concepts like colors, moods, or styles. This makes image descriptions more detailed and better reflects their content.

An example of a tag-based recommendation algorithm
Let's take a closer look at one algorithm - tag-based recommendations. Here's how it might work step by step.
- Describing products with tags - each product (e.g., clothing, film, book) is described with a set of tags or keywords. These can define the category ("dress," "comedy"), features ("cotton," "romantic"), or product purpose ("summer," "evening"). Tags can be added manually or automatically based on image analysis.
- Collecting data about your preferences - the system tracks what products you view, purchase, and rate. Based on the tags of these products, it creates your preference profile.
- Determining preference strength - the system analyzes your actions: what you view, purchase, how long, and how often. If you usually choose products described with a specific tag, the AI assumes that tag interests you. The more often you buy products with a given tag, the greater its weight in your profile.
- Filtering tags - the system removes tags that are less relevant to you, i.e., those appearing in products you only glanced at.
- Finding matching products - the AI searches the product database and finds those that best match your preference profile, meaning those with the most tags you prefer.
- Presenting recommendations - selected items are presented to you as "products chosen especially for you." Often, the system explains why it thinks you might like them (e.g., "similar to recently viewed").
- Learning from reactions - artificial intelligence observes how you respond to recommendations, such as whether you click on suggested products, how long you view them, and if you buy them. Based on this, it updates your preference profile and adjusts future suggestions.
Tag-based recommendations involve a complex process that requires continuous data collection, analysis, and learning from your reactions. All this to predict as accurately as possible what you might like and make finding the perfect products easier.

Ready-made tools for automatic tagging
Imagga is a service that uses artificial intelligence to analyze images. Just upload a product photo, and it will return a list of matching tags. The tool can also recognize more complex and subtle aspects such as dominant colors, moods, or types of activities depicted.
Thanks to such tools, creating product descriptions for recommendation systems is much simpler and faster. Online stores can easily tag their products, translating into more accurate recommendations for users.
Recommendation systems are powerful tools that can accurately predict what we will like using artificial intelligence. Their operation is based on advanced machine learning algorithms, analyzing vast amounts of data, and automatic image recognition.
It's worth understanding how they work. This way, you'll better understand where the suggested content comes from. Arm yourself with knowledge to use the benefits of artificial intelligence more consciously in everyday life.