How content is created and distributed at Linkedin. The role of follows and follow recommendations. (4 types of follow recommendations)
3 major problems to address:
- Personalization requires operating at a very large scale: matching members to other members is a quadratic problem on the order of 10^17 pairs.
- Two-step optimization problem: follow recommendations have to optimize for likelihood of a follow edge being created, but also have to optimize for actual engagement that can result from this edge over time. Added complication of different models that rank content once it’s generated that are always changing.
- Dynamic follow recommendations: How to be responsive to real-time actions that the member is taking, while using resources efficiently and operating at a massive scale.