An Industrial-Scale Sequential Recommender for LinkedIn Feed Ranking
LinkedIn just deployed a massive new transformer model to rank your feed.
Deep Dive
LinkedIn has replaced its old ranking model with a new transformer-based sequential recommender called Feed-SR. The model, detailed in a new research paper, is now the primary system for LinkedIn's main feed. In online A/B tests, it delivered a significant 2.10% increase in member time spent compared to the previous production model. The team chose this architecture over other sequential and LLM-based options for its optimal balance of performance and production efficiency.
Why It Matters
This shows how major platforms are moving beyond simple models to complex, engagement-boosting AI that directly shapes what you see.