Recommender systems | Revenue & Trends

The hottest Substack posts of Recommender systems

And their main takeaways

Self-attention helps summarize a list of information, making it easier to find what's most relevant, like recent videos you watched.
Graph attention looks at how items in a network relate to each other, like understanding social connections in a network.
Target-aware attention checks how relevant certain items are based on your past choices or queries, helping improve recommendations.

Scalability is key for embedding-based recommendation systems, especially when dealing with billions of users. Finding effective ways to limit the search can help manage this challenge.
It’s important to deliver value not just to viewers but also to the recommended targets, as this can improve user retention. Balancing recommendations for both sides can create a better experience.
Using advanced algorithms can help ensure viewers don’t get overwhelmed with too many recommendations while also making sure that every target gets the attention they need. This balance is crucial for effective recommendations.

Friend recommendation systems use connections like 'friends of friends' to suggest new friends. This is a common way to make sure suggestions are relevant.
Two Tower models are a new approach that enhances friend recommendations by learning from user interactions and focusing on the most meaningful connections.
Using methods like weighted paths and embeddings can improve recommendation accuracy. These techniques help to understand user relationships better and avoid common pitfalls in recommendations.

Training an early ranker to mimic the final ranker can improve top-line metrics and reduce costs
Knowledge distillation involves training a student model, the early ranker, to learn from a teacher model, the final ranker
Implementing knowledge distillation through shared or auxiliary tasks can increase alignment between the early and final rankers

Reducing selection bias and popularity bias in ranking is important for recommender systems.
An advocated approach is to factorize user interaction signals to account for biases originating from power users and power items.
The proposals for causal/debiased ranking involve factorization, mutual information, and mixture of logits to improve the ranking model.

Get a weekly roundup of the best Substack posts, by hacker news affinity: