Content Recommendation for a National Newspaper
A national newspaper deployed a hybrid content recommendation engine balancing relevance and discovery, optimizing for reader lifetime value rather than clicks, with editorial override capabilities.
The challenge
All readers saw the same content feed regardless of interests. No personalization existed. Reader retention (LTV) was declining. The editorial team wanted to maintain control alongside algorithmic personalization. Click-through rates, the primary metric, did not predict subscription renewal.
What we built
We built a hybrid recommendation engine combining collaborative filtering, content-based embeddings, and a ranking layer that balances relevance and diversity. User behavior modeling captured views, reading time, and shares. Batch model training with near real-time serving provided up-to-date recommendations. Editorial override capabilities let the team pin, boost, or suppress specific content.
Outcomes
Content feeds optimized for each individual reader
Readers encounter familiar and novel content
Models trained on retention, not just click-through rates
From data analysis to production recommendation engine
“Our readers discover more stories they care about, and they stay longer.”