Optimize the process for classifying social media mentions as spam/feedback/neutral/news. The client was handling 36K items per day with an outdated model and sending an additional 19K items per day to Toloka for manual labeling.
Solution
A human-in-the-loop workflow using Toloka's Adaptive ML Models was used to train an accurate model. With improved accuracy using the new model, manual labeling was reduced from 19K to 9K items per day.