Table of Contents

1. Problem Definition
2. Methods
3. Impact
4. TODO & Questions & Further Reading

Problem Definition

1. We covered three aspects beyond the choice of algorithm:
   1. Modelling
   2. Experimentation
   3. Serving
2. In this paper
   1. challenges and solutions
   2. display an overview of the business gains of the step changes in the Machine Learning ranking of Booking.com.
3. Main conclusion
   1. Our main conclusion is that there are multiple aspects that need to be carefully addressed for the creation of a Machine Learned ranker in a large-scale commercial setting.

      Methods

4. Modelling
   1. Signals definition
   2. Feature representation
   3. Bias
5. Experimentation
   1. Leakage: retraining and features
   2. Speed of experiment: interleaving
6. Serving

Impact

1. Improvements of each step: refer to plot

TODO & Questions & Further Reading

1. re-training leakage -> Isolated feedback loops of data -> why & potential influence ??
2. can borrow some feature from feature representation
3. signal definition -> useful

• Previous
  2019 - Deep Learning based Dynamic Pricing Model for Hotel Revenue Management