Two-Tower Model
Definition
Two-tower models compute query and possible matches separately for fast comparison.
An architecture with separate encoders for query and candidate, whose vectors are efficiently compared.
Simply explained
One side describes the search query, the other the possible offers. Their vector representations are compared to find suitable matches.
Why it matters in practice
The term improves reliable access to data and knowledge as well as their usable linkage in digital systems.
Example from business
A B2B shop pre-indexes product vectors and computes only the query vector online.
Limits and pitfalls
- Separate encoders model fine interactions less strongly than cross-encoders; negatives and index freshness are critical.
Origin & history
Two-tower architectures became widespread for large-scale recommendations and retrieval in the 2010s.
Use in mid-sized companies
- Possible application: A B2B shop pre-indexes product vectors and computes only the query vector online.
Frequently asked questions
What does two-tower model mean?
An architecture with separate encoders for query and candidate, whose vectors are efficiently compared. What should you watch out for? Separate encoders model fine interactions less strongly than cross-encoders; negatives and index freshness are critical.
Sources
- Recommending movies: retrieval – getrennte Query- und Candidate-Modelle sowie vorab berechnete Kandidatenrepräsentationen. · TensorFlow Recommenders: Recommending movies: retrieval – getrennte Query- und Candidate-Modelle sowie vorab berechnete Kandidatenrepräsentationen.
- Deep Neural Networks for YouTube Recommendations. · Paul Covington, Jay Adams und Emre Sargin (2016): Deep Neural Networks for YouTube Recommendations.
Editorially reviewed on: