Combining Neural Networks and Regression Tree for Dynamic Pricing in Mobile Advertising
Technical level: Advanced
In a diversified mobile advertising marketplace, it is important to dynamically set the minimum CPM bid in order to maximize revenue as well as meeting CPM expectations. Our approach combines neural networks and a customized regression tree for accurate prediction and effective subsidization. The dNN model predicts revenue and impressions as functions of the minimum CPM bid and other supply/demand features. The regression tree model iteratively selects the most important features for the prediction functions. This generates a highly compact prediction table that allows us to do dynamic pricing for the whole marketplace.
- The problem of dynamic pricing and challenges
- Prediction with dNN
- Prediction aggregation with regression tree
- Evaluation results
Basic understanding of neural networks and tree based models
Wei Li has a Ph.D. degree in Computer Science from University of Massachusetts, Amherst. She is now a senior research scientist at InMobi, focusing on marketplace optimization. Before that, she has worked on contextual advertising in Yahoo and Google.
Rajiv Bhat has a Ph.D. degree in Theoretical Physics from University of Colorado at Boulder. He is now the Senior Vice President of Data Sciences at InMobi. Before that, he co-founded Mertado, a social shopping platform that was bought over by Groupon.
Neural-network Field Aware Factorisation Machines for Online-behaviour Prediction
In the AdTech mobile-app industry, bidding for each and every ad-request at a suitable price and in real-time is absolutely critical. Thus, there is always a need for more scalable and more accurate prediction models, which drive higher revenue. more