Brand Recommendation
Sweetspot
Problem
- Lack of a model that recommends suitable brands for vacant units by considering both commercial district and brand characteristics
- Existing approaches only consider simple factors like purchasing power and brand power, ignoring complex interactions between district features (e.g., foot traffic) and brand-specific attributes
Solution
- We perform sales prediction modeling on annual sales data from brand franchise stores
- Models based on DNN and GNN that consider non-linearities such as interaction terms between commercial district characteristics and brand characteristics showed better performance as expected
Contributions
- Problem definition and preprocessing raw data into a format suitable for modeling
- Implementation of multiple models including HinSage-based GNN model