Abstract
Deliver an E2E CTR (Click through rates) Prediction Model using Deep Learning techniques to improve Outbrain existing Recommendations. The Project involves building a working pipeline with Classic Machine Learning techniques, to improve it with DL using Tensorflow/Keras and to evaluate the models with Outbrain Metrics.
Challenges
- Working with large amount of correlated data
- Setting up and running a Pipeline on a remote GPU
- Testing and fine-tuning state of the art optimization methods
Achievements (according to KPIs):
- Improved the model in production by 30% in RMSE
- Delivered a working pipeline
- Found useful insights on how DL methods behave on Outbrain Data
Further development
- Improve the model performances in terms of prediction time
- Write the model in production code
- A/B test the model performances online