Search⌘ K
Home
Courses
Grokking the Machine Learning Interview

Ranking

Explore how to rank media content effectively in recommendation systems by predicting user watch probabilities. Understand when to apply logistic regression and tree ensembles versus deep neural networks with sparse and dense features. Learn techniques like embedding historical user interactions and re-ranking for diversity to optimize personalized recommendations.

We'll cover the following...

The ranking model takes the top candidates from multiple sources of candidate generation that we have discussed. Then, an ensemble of all of these candidates is created, and the candidates are ranked with respect to the chance of the user watching that video content.

Here, your goal is to rank the content based on the probability of a user watching a media given a user and a candidate media, i.e., P(watch|(User, Media)).

Ensemble of candidate generation models
Ensemble of candidate generation models

There are a few ways in which you can try to predict the probability of watch. It would make sense to first try a simplistic approach to see how far you can go and then apply complex modelling approaches to further optimize the system.

First, we will discuss some approaches using logistic regression or tree ensemble methods and then a deep learning model with dense and sparse features.

Deep learning should be able to learn through sparse features and outperform simplistic approach. Still, as we discussed in earlier problems, generalization with a deep NN model needs an order of ...