Most of us use Google, Netflix, Spotify, or other platforms that provide us with recommendations.
Have you ever noticed that when you look for a new book, movie, or restaurant, the search results often suggest things you’ve already tried or bought?
This is because the artificial intelligence (AI) tools these platforms use can trap us in a ‘filter bubble’, where we only see stuff similar to what we have previously liked or bought.
But now, a group of computer scientists has created a new way to break free from these filter bubbles.
They developed an algorithm called ‘Pyrorank’ that acts like it’s part of the natural world.
Think about how birds find food. They don’t all go to the same spot; they explore different areas. The scientists used this concept to make Pyrorank provide more diverse and interesting results.
Anasse Bari, one of the creators of Pyrorank and a professor at NYU’s Courant Institute of Mathematical Sciences, explained that the natural world is an excellent place to find inspiration for solving tech problems.
He also pointed out that the traditional systems of giving recommendations work based on similarity. For example, if you bought an Apple product, you would keep seeing more Apple products in your recommendations.
However, these traditional systems have limitations. They might keep showing you the same type of content again and again, which can lead to problems.
For instance, people may only see news that aligns with their views, which doesn’t expose them to different perspectives.
To solve these issues, Bari and his team created Pyrorank. It considers what the user is looking for, and provides a variety of suggestions, but tones down the influence of the user’s past interactions.
The cool thing is, Pyrorank can be added to existing systems, making it easier to use and saving time for engineers.
When tested, Pyrorank produced a wider variety of recommendations compared to traditional systems. The catch is, it might slightly reduce how accurately it can predict what the user will like. But according to the researchers, the balance between diversity and accuracy can be adjusted depending on the situation.
Bari concludes that the best recommendation systems should understand and reduce biases, which will lead to more effective recommendations and healthier use of these platforms.
Adding diversity to recommendations is a crucial step toward overcoming the limitations of current systems.
The study was published in Advances in Swarm Intelligence.
Follow us on Twitter for more articles about this topic.
