Hierarchical segmentation model of Kijiji sellers based on site activity using scikit-learn. A separate clustering was performed on each category to eliminate cross-category confusion and reduce computational load.
Model features:
- Email domain(public or private)
- Ads posted/day
- Replies sent/day
- Buy:Sell ratio
- % Ads posted per L2 category
- % Replies sent per L2 category
- Avg. ad duration
- Total inventory value
- Avg. price/ad
- Avg. feature spend/ad
- Replies received/ad
- Connections/ad
- Avg. conversation depth
Model pipeline:
Five main seller types were identified across categories:
