GrIS paper reframes Semantic IDs as recursive graph partitioning for generative recommendation
_reachsumit · x · 2026-10-02
Core idea
Semantic ID construction in generative recommendation can be reframed as a recursive clustering problem: hierarchically partition a graph whose nodes carry semantic content and whose edges carry collaborative signal. The resulting framework, GrIS, subsumes rather than displaces prior methods.
Key findings
- RQ-VAE and RQ-KMeans are recovered as the special case of an empty graph — content-only quantization is one corner of a larger design space along two previously collapsed axes: graph construction and recursive partition algorithm.
- Two instantiations: RecDMoN (hierarchical assignment via differentiable graph pooling) and RQ-GAE (RQ-VAE extended with graph-aware item representations and a graph reconstruction objective).
- On multiple real-world datasets, GrIS consistently outperforms content-only CF baselines.
More from Research
- New video traces how adversarial objectives evolved beyond GANs and self-play — manicman1999 · 2026-10-02
- Peer-reviewed paper questions whether gene expression model benchmarks are informative — simocristea · 2026-10-02
- Oxford VGG Unveils SynCity 3000, Generating Globally Coherent Scene-Scale 3D Worlds — rsasaki0109 · 2026-10-02
- Christian Szegedy revisits 2019 interview: his 'crazy' timelines for AI math and code — ChrSzegedy · 2026-10-02
- Omni-Embed-Mini: A 0.9B Embedder Adds Five Modalities Without Touching Text Weights — _reachsumit · 2026-10-02
- Apple paper: structured selection-based reasoning cuts search agent latency by 90% — _reachsumit · 2026-10-02