Collaborative filtering

Collaborative filtering on Instagram finds possible recommendations by comparing patterns in different people’s interests and activity1. As Adam Mosseri explained in June 2021, Instagram starts with posts someone has liked and the accounts that shared them, then looks at what else people with overlapping likes or follows enjoy1. Meta’s patents illustrate related approaches for identifying possible page and location recommendations, rather than establishing how Instagram currently recommends posts2 3. The term covers multiple techniques: Meta’s April 2026 research compares “classic,” “neural” and “generative” forms4. In the latest account discussed here, Meta’s engineering team explained in August 2026 that information about an ad’s content complements shared-interaction signals, especially for new ads or advertisers with little engagement history—a limitation and response described for advertising, not guidance on creators’ organic reach5.

Collaborative filtering on Instagram and in Meta’s ad research

Meta research and patents. These may not describe how Instagram works today.

Instagram’s collaborative filtering starts with posts a person has liked and the accounts that shared them, according to Mosseri1. Instagram then looks for other people who liked those posts or follow those accounts, and examines what else those people like and whom they follow1.

For advertising, Meta’s April 2026 research describes a system deployed for ad recommendations to Facebook and Instagram users and presents several forms of collaborative filtering as comparison methods4. Meta’s paper labels these forms “classic,” “neural” and “generative”4. Meta’s paper compares them alongside three other families of methods for finding possible recommendations4.

Collaborative filtering on Instagram and in Meta’s ad research

See also

References