Instagram Ads Ranking and Delivery Skills: Machine Learning

Built innovative machine learning models that can predict if a user will click/convert on a given advertisement utilizing state-of-art methods such as large linear models, collaborative filtering, factorization machines, and deep learning. Bagged significant offline NE and AUC gains, and online A/B test clicks and conversion gains. Optimized ads ranking models and ranking settings across multi-stage ranking setup for overall improved precision and recall. Defined key metrics for multi-stage consistency, and improved it by 15+% via systematic tuning of models across stages. Built machine learning models using state-of-art techniques that can predict if a user will like/ dislike a given advertisement. These models are utilized in online ad auctions to dynamically trade-off between ads quality versus ads revenue. Helped reduce revenue investment in ads quality by 20+% while maintaining steady ads like and dislike rates. Architected and implemented an end-to-end scalable framework that allows building and serving sparse features used in ads ranking via a simple configuration. Reduced development time for a sparse feature from several days to several minutes. Doubled the number of active sparse features used in ads ranking. Designed a novel machine learning framework that can predict personalized future long-term product engagement for a user. Utilized this framework to help make product launch decisions, and for understanding dynamic trade-off between long-term engagement versus short-term engagement and advertisement revenue. Skills: Machine Learning