Russia has found a way to speed up and diversify network recommendations

Scientists from T-Bank AI Research have developed a method that allows for faster and more diverse recommendations on online platforms. The method, called Sampled Maximum Marginal Relevance (SMMR), helps to create more personalized collections tailored to the user's interests, without focusing on one type of content, T-Bank said.

The results of the study were presented at one of the leading conferences in the field of machine learning and AI, ACM SIGIR, which takes place in Padua, Italy, on July 13-18.

Traditional algorithms tend to select the most suitable objects — goods, movies, news — based on user preferences. However, this approach creates a so-called "information bubble" when the user sees only those products or content that are similar to his previous interests, the company explained.

The SMMR method solves this problem through probabilistic selection: the algorithm randomly selects from a limited range of suitable options.

In addition, SMMR is faster than the well—known analogies - MMR (Maximum Marginal Relevance) and DPP (Determinantal Point Process). The new method turned out to be 2-10 times faster and at the same time provided a 5-10% increase in the variety of recommendations. This opens up new opportunities for the largest fintech companies, media and social networks, online cinemas, marketplaces and other online platforms, as the services will become more convenient and interesting for users, according to T-Bank. The company plans to introduce the new method into its own digital services to improve the quality of recommendations, T-Bank emphasized.