BACO compresses recommender embedding tables by over 75% using balanced co-clustering of users and items that maximizes intra-cluster connectivity while enforcing size balance, with at most 1.85% recall drop and up to 346X speedup over baselines.
However, their approach simply applies modularity maximization, without ade- quately considering the unique data characteristics and clustering biases of recommender systems
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.IR 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Balanced Co-Clustering of Users and Items for Embedding Table Compression in Recommender Systems
BACO compresses recommender embedding tables by over 75% using balanced co-clustering of users and items that maximizes intra-cluster connectivity while enforcing size balance, with at most 1.85% recall drop and up to 346X speedup over baselines.