Knows You Better: Russia Develops New SMMR Algorithm for Smarter Content Recommendations
From movies and music to products you didn’t know you wanted—Russia’s new SMMR algorithm promises smarter, faster, and more diverse recommendations.
New Capabilities
Developed by researchers from T-Bank AI Research, MIPT, and HSE University, the Sampled Maximal Marginal Relevance (SMMR) method generates recommendation sets up to 10% faster and with greater diversity compared to traditional approaches.
Conventional recommendation engines prioritize the most relevant items based on user history. The result? Monotony. SMMR breaks that loop. It doesn’t just select top matches—it randomly chooses from a curated set of appropriate options. This blend of personalization and randomness introduces users to new content they might never have discovered.
The algorithm’s code is already available on GitHub. The team presented their findings at the 48th ACM SIGIR International Conference on Information Retrieval, currently taking place in Padua, Italy.
More Variety, More Engagement
SMMR opens up fresh possibilities for fintech firms, streaming services, online marketplaces, and social media platforms. By introducing a touch of surprise into recommendations, the algorithm enhances user engagement and expands discovery.
For example, a standard algorithm might suggest 10 books from the same genre. SMMR, by contrast, could add two or three titles from entirely different genres—still relevant, but unexpected. Sellers benefit too: more products get visibility, and consumer experience becomes richer and more personal.
Tested and Ready to Deploy
The algorithm requires no language adaptation, making it plug-and-play for Russian platforms. SMMR also complies with current national regulations. According to T-Bank, the company will integrate the method into its own digital services. In its T-Shopping platform, SMMR will be used to enrich product suggestions, while in its social app Pulse, it will increase feed diversity.
For Russia’s tech industry, SMMR is a meaningful step toward digital autonomy. It offers a credible alternative to Western solutions, expands access to diverse content, and strengthens national tech infrastructure. On the export front, the algorithm has already shown solid performance on open datasets like MovieLens (movies), Dunnhumby (shopping), and MIND (news). T-Bank reports that it works well in both static and fast-changing environments, such as news feeds.
Global Relevance
SMMR aligns with global trends in personalized content delivery. Retailers, streaming services, and digital platforms worldwide are doubling down on smart recommendation systems to improve customer loyalty and drive revenue.
Amazon leverages purchase and search history. Netflix factors in watch history and ratings. Spotify crafts playlists based on personal listening habits. In Russia, Yandex is integrating generative models into its recommender engines. SMMR combines speed and quality, with monetization projected within 1–2 years.
