Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Sep 7·5 min read
Beyond Co-purchase Relation: Evolution of Complementary Recommendations at Allegro
TL;DR
AlleCompanion is a new retrieval framework that enhances complementary product recommendations by filtering out irrelevant co-purchase data, leading to improved user satisfaction and increased revenue. It effectively serves over 20 million users monthly, demonstrating significant growth in attributed gross merchandise value (GMV).
✦ Why It Matters
Implement category-constrained recommendation systems to enhance user experience and drive sales effectively.
Key Takeaways
- Utilize category constraints to improve recommendation relevance.
- Incorporate user feedback for continuous model refinement.
- Leverage multi-source data for better insights into user behavior.
- Focus on enhancing both organic and sponsored product placements.
How It Works
AlleCompanion employs a Two Tower architecture that separates the representation of products and user behavior. The Category Adapter constrains the embedding space to ensure that only logically complementary products are suggested.
ComCat acts as a translational layer, synthesizing insights from various data sources, including expert rules and user feedback, to filter out noise from co-purchase data.