Rebuilding Linkedins Follows Recommendations With Llm Based Semantic Retrieval And Ranking

LinkedIn's redesigned follow recommendations use profile meaning to find creators for members with little activity history. The system converts profile descriptions into numerical representations trained to bring likely member–creator matches closer together. It searches those representations for candidates and reuses them when ranking recommendations.

This semantic matching runs alongside existing popularity and connection-based methods. It gives the system another way to surface a relevant creator whose audience is small. LinkedIn reports increased follow rates in production comparisons, with the strongest gains among new members, but provides no numerical effect size in the article text. System design and reported results.