Online Display Advertising Optimization with H2O

by Hassan Namarvar, Principal Data Scientist

We held a data science meetup (as a series of SF data mining meetups) at ShareThis headquarter at Palo Alto on Dec 9th, 2014. I presented our teamwork about online display advertising optimization. In online display advertising, the ultimate goal is to provide a best and relevant ad to an online user so that to influence him/her to take an action such as purchasing a product or signing up for a service. This requires estimating the probability of conversion for a given user, content, advertiser, location, device and so on. Conversion estimation is extremely challenging task since conversion events are rare events and data dimension is huge.  In this talk, I presented how, at ShareThis, we tackled conversion estimation problem. More specifically, I described how we built CPA models by leveraging ShareThis social media and Ad exchange datasets and applying the state-of-the-art machine learning algorithms such as GLM, GBM, and Random Forest provided by the H2O platform. I presented some results from live advertising campaigns at production to show the effectiveness of our approach. For more details, you can watch the following video and deck from yesterday talk