OMEGA adds collaborative memory to generative recommendation

_reachsumit · x · 2026-08-04

This paper proposes OMEGA, a collaborative memory augmentation framework for generative recommendation.

The method compresses user behavior into compact representations, stores them in a collaborative memory bank, and retrieves relevant memories with a target-aware mechanism. A gated cross-attention module then fuses the retrieved memories with local user context to improve generative recommendation quality on multiple datasets.

Original post →

More from Research

Research channel →