Realtime Recommendation at Blendle

Summary

This presentation examines the challenges and solutions for implementing real-time personalized content recommendation systems in digital journalism platforms, specifically addressing the unique constraints of news article recommendation compared to traditional domains like music or movies. The work tackles critical issues including the short shelf-life of news content, cold start problems where collaborative filtering fails, and the need for content diversity while leveraging implicit user feedback through reading behavior and explicit signals like refunds and likes. The research contributes practical insights into building recommendation systems for micropayment journalism platforms, demonstrating how to balance personalization with content discovery in a domain characterized by high content turnover, diverse user preferences, and limited user interaction data compared to entertainment platforms.