Rebuilding LinkedIn’s Follows Recommendations with LLM-Based Semantic Retrieval and Ranking

LinkedIn rebuilt its Follows Recommendation system, replacing heuristic candidate generation and popularity-driven ranking with LLM-based semantic retrieval. Member and creator profiles are converted into narrative prompts, encoded into embeddings via a fine-tuned bi-encoder trained with contrastive learning (InfoNCE loss, LoRA, mixed precision, gradient checkpointing), and used for both offline (FAISS, Ray) and online (Proxima-hosted) retrieval. The same embeddings feed the downstream L2 ranking model via learned dimensionality-reduction projections. Production A/B tests showed statistically significant follow-rate lifts, especially for new members with sparse activity history.

Source: https://www.linkedin.com/blog/engineering/ai/rebuilding-linkedins-follows-recommendations-with-llm-based-semantic-retrieval-and-ranking. 8 Sync News only summarizes and links out; content copyright belongs to the authors and original sources.