Elena Lebedeva, ML developer, VK Reklama (VK Ads, contextual targeting)
A talk about how the team built a dataset for multilabel classification of ad texts, why annotator labeling did not work, and how the labeling was replaced with an LLM. The model itself did not change: the metric gain came only from better labels.
Continue reading — it's free
Google or a one-time email code. No password.