A team of computer scientists from New York University, including a Moroccan, has designed a new algorithm called “Pyrorank” that reduces the impact of user profiles and broadens recommendations.
“Pyrorank” generates the recommendations that always reflect the objective of the research, producing more diversified and useful results, underlines the American university in a press release.
“When it comes to inspiration for finding solutions to computer problems, nature is the perfect place to look,” says Moroccan Anasse Bari, associate professor of computer science at New York University and co-creator of the algorithm.
“Natural phenomena, such as the flight of birds in search of food, show that nature can often find optimal, yet simple, solutions to meet needs,” he added.
This is because the AI tools that many companies rely on push users into a “filter bubble,” resulting in recommendations that are the same or very similar to what was purchased previously.
Recommender systems, used by Google, Netflix, and Spotify, among others, are algorithms that use data to suggest or recommend products or choices to consumers based on past purchases, search history, and demographics. users.
However, these parameters skew search results because they place users in filter bubbles.
“The traditional way recommender systems work is to base recommendations on the notion of similarity,” says Bari, who leads the Predictive Analytics and AI Research Lab at the Courant Institute of Mathematical Sciences at the University of New York.
“This means that you will see similar results in choice lists and recommendations based on users similar to you or items similar to those you have purchased”, adds the Moroccan computer scientist, author of this algorithm in collaboration with Nicholas Greenquist and Doruk Kilitcioglu.
The limitations of existing recommender systems have become starkly evident, New York University points out.
To address these concerns, Bari and his colleagues created Pyrorank, an algorithm that takes into account the content a user is looking for by capturing an array of recommendations while decreasing the importance of what the user has already purchased or with. which he interacted.
Pyrorank works as an algorithmic “add-on” to the recommender systems available.
In testing the viability of the algorithm, the researchers compared the search results generated by the Pyrorank add-on with those of traditional recommendation systems using three large datasets: MovieLens, which offers movie ratings generated by user, as well as Good Books and Goodreads, which host readers’ book notes.
They then ran a series of experiments to determine which systems created more diversity in recommended content while staying true to the goals of the core recommendations.
Overall, systems using Pyrorank generated more diverse recommendations than existing ones, points out the American university.