SOMIN published a paper titled Drift-Aware Multimodal User Representation Learning via Multi-Scale Temporal Modeling and Sparse Mixture-of-Experts titled DUMoE, which tackles user preference as a trajectory and introduces a framework for temporal drift modeling.
Published
Signal category
Research & Knowledge
Quote
“There is a paper that came out of arXiv this week that I think deserves more attention than it will probably get outside the recommender systems community. The work, titled "Drift-Aware Multimodal User Representation Learning via Multi-Scale Temporal Modeling and Sparse Mixture-of-Experts" (DUMoE), tackles something that has been quietly frustrating practitioners in marketing AI for years: the fact that user preference is not a static vector, it is a trajectory.”
— Prof. Aleks Farseev|SOMIN team
Company
SOMIN
AI-powered creative intelligence platform that decodes how brands communicate - turning billions of ads into insights.
- Industry
- Desktop Computing Software Products
- Location
- London, GB
- Company size
- 28 employees
SOMIN is an AI-powered brand strategy and creative ideation platform that helps agencies develop stronger pitch ideas and deliver more insightful QBR and annual reporting, while enabling brands to uncover competitor blind spots and generate hyper-local communication ideas grounded in real market signals. By analysing billions of competitor ads, posts, and campaign patterns, SOMIN turns fragmented market noise into actionable strategic opportunities. Using semiotic analysis across large-scale competitive communication, the platform identifies underused emotional territories, overlooked cultural narratives, and whitespace opportunities that competitors have not yet claimed. This allows agencies to pitch with fresher, more evidence-based ideas, and helps brands build locally relevant messaging that stands out in crowded markets.
Founded 2018