Design People You May Know
People You May Know (PYMK) is a recommendation system that suggests potential connections to users of a social or professional network. It predicts which relationships are likely to form, using signals such as mutual connections, shared schools or employers, and past interactions, and it has to account for both the person seeing the suggestion and the person being suggested.
Imagine opening LinkedIn and seeing a section called People You May Know. It recommends a former colleague, a university classmate, or someone who works in your industry. Sometimes, these suggestions are surprisingly accurate.
But how does LinkedIn decide which people to recommend out of hundreds of millions of possible profiles? And how can it make those recommendations personalized, relevant, and fast?
This is a classic Machine Learning System Design problem, and it's more interesting than it initially appears. At its core, we're predicting which relationships are likely to form in a social network. But a production system needs to solve several additional challenges. We have to find relevant candidates without searching the entire network, predict whether two people will actually connect, account for both people's preferences, and continuously improve recommendations as the network evolves.
In this article, we'll design a People You May Know system from scratch, following the same approach we would use in a real Machine Learning Engineer interview. We'll cover requirements, metrics, candidate generation, feature engineering, model selection, training, evaluation, serving infrastructure, and production challenges.