Page Summary

outlined_flag

• Recommendation systems often use a three-stage architecture: candidate generation, scoring, and re-ranking.

• Candidate generation narrows down a large pool of potential recommendations to a smaller subset for further evaluation.

• Scoring assigns relevance scores to the candidates and ranks them to identify the top recommendations.

• Re-ranking adjusts the initial ranking to address additional factors like user preferences, diversity, and content freshness.

One common architecture for recommendation systems consists of thefollowing components:

• candidate generation
• scoring
• re-ranking

Candidate generation

In this first stage, the system starts from a potentially huge corpus andgenerates a much smaller subset of candidates. For example, the candidategenerator in YouTube reduces billions of videos down to hundreds or thousands.The model needs to evaluate queries quickly given the enormous size of thecorpus. A given model may provide multiple candidate generators, each nominatinga different subset of candidates.

Scoring

Next, another model scores and ranks the candidates in order to selectthe set of items (on the order of 10) to display to the user. Since thismodel evaluates a relatively small subset of items, the system can usea more precise model relying on additional queries.

Re-ranking

Finally, the system must take into account additional constraints for thefinal ranking. For example, the system removes items that the userexplicitly disliked or boosts the score of fresher content. Re-rankingcan also help ensure diversity, freshness, and fairness.

We will discuss each of these stages over the course of the class andgive examples from different recommendation systems, such as YouTube.