VKontakte Taught Its Feed to Distinguish Content Interest from Purchase Intent

💻 VKontakte Taught Its Feed to Distinguish Content Interest from Purchase Intent

VK AI engineers created a separate product circuit in the Feed: candidate selectors (a catalog of about 74,000 product posts, first partner — Ozon) and a multi-target ranker that evaluates the probability of an order instead of CTR. Logging of product events was raised from the standard 2% to 100% via the VK Discovery platform.

🌍 A working playbook: isolate a new domain in the recommender, first address the data deficit, then shift the target from proxy events to the target — the order. Useful for services with rare target events and cold start.

👤 A separate model now selects "shops," targeting not likes and CTR, but the probability of leading you to an order. Over five months: end-to-end order conversion +53%, orders ×43.2, partner transitions ×55.3, CTR +95.6%.

Source 1: https://habr.com/ru/companies/vk/articles/1088034/