Notebook Click Model: Baseline, Model Comparison and Follow-Ups
Problem statement
How would you build a notebook that predicts clicks on a content app? You have impression records with post identifiers, topic activity, the displayed topic and click outcomes. Posts repeat, data may be messy, and activity must come from before display. Timestamps and user identifiers are missing, making deployment validation harder.
Build the notebook, calculate its constant baseline loss and choose features that capture how topic preference changes with activity.
Show hint
What information would you need to reproduce a prediction made at display time?
Relevant roles: ML Engineer, Data Scientist and Applied Scientist.
Related free problems
- Position Bias in a Click-Trained Feed RankerML System DesignHard