Method

Pseudo-labeling

Training on model-generated labels; semi-supervised learning.

10 cases from 9 companies apply Pseudo-labeling. 1 was checked in an online A/B test; 6 report results in production.

Typical tasks: Rank search results, Label data with LLMs, Reduce ML cost, Catch fraud and abuse and 1 more. Companies: Yandex, X5 Tech, VTB, Avito, Pinterest, VK and 3 more. Often combined with: Threshold selection, Few-shot prompting, LLM labeling, Gradient boosting, Language-model pretraining and 35 more.