RP / 2026 In progress

Research proposal · Generative recommendation

Mixed-Curvature Semantic IDs

Preserving heterogeneous local topology through RQ-VAE quantization

Generative recommenders compress item semantics into discrete Semantic IDs. This proposal asks what that compression does to the item graph — and whether curvature can be allocated, rather than chosen once, to keep the local structures that matter.

  • RQ-VAE
  • Semantic ID
  • Mixed curvature
  • Graphlets

Working note · November 2026 Wenyu Liang · Monash University

Three local graphlet structures paired with three geometric biases A triangle is paired with a Euclidean grid, a star with a hyperbolic fan, and a chain with a spherical orb. The figure asks which geometry fits which local neighbourhood of the same item. LOCAL STRUCTURE GEOMETRIC BIAS Triangle closure Euclidean Star hub Hyperbolic Chain ordering Spherical One item. Three neighbourhoods. One codebook.
The same item can sit in a triangle, a star, and a chain. The proposal treats geometry as a per-item, per-level allocation rather than a single global choice.

The quantization is not topology-neutral

RQ-VAE is usually trained on item semantics and judged by reconstruction, codebook usage, or downstream accuracy. The item graph it leaves behind is rarely the object of study.

In preliminary experiments, a semantic-only RQ-VAE still recovered part of the real item–item behavior graph after quantization. The recovery was not uniform: triangle, star, and chain neighborhoods were preserved to different degrees, and their retention rates correlated differently with downstream recommendation performance.

That unevenness is the starting point. If one item can belong to a triangle, a star, and a chain at the same time, a single geometry and a uniform reconstruction objective may not preserve those structures equally.

Closure

Triangle

Tightly coupled substitutes or complements. Does quantization preserve the closure, or collapse the three items into one region?

Concentration

Star

A hub with many leaves. Once codes are shared, is the hub retained, its leaves, or neither?

Ordering

Chain

A graded or sequential relation. Does a discrete code preserve the order, or only proximity?

Working hypothesis

An item participates in several local topologies at once. A single Euclidean codebook and a uniform reconstruction objective cannot preserve them equally. Geometry should become an allocatable resource: hyperbolic where hierarchy dominates, spherical where closure or cycles dominate, Euclidean as the stable default.

Three questions, one fixed budget

RQ1

Which topologies survive?

How does RQ-VAE quantization reshape triangle, star, and chain structure, and which of these changes track downstream recommendation utility?

Diagnose
RQ2

Can curvature be allocated?

Can Euclidean, hyperbolic, and spherical biases be assigned per item and per codebook level to match heterogeneous local topology?

Model
RQ3

What is worth preserving?

Under a fixed Semantic ID budget, does selective structure preservation outperform uniform neighborhood reconstruction?

Constrain

A topology-aware quantization stack

01 / Probe

Probe the graph

Build a KNN graph on pre-quantization item embeddings. Extract graphlet signatures and neighborhood descriptors, then measure the same structures after quantization.

02 / Allocate

Allocate curvature

Learn a per-item, per-level mixture over Euclidean, hyperbolic, and spherical components. Hyperbolic capacity for hierarchy, spherical for closure and cycles, Euclidean as the default.

03 / Constrain

Select what to preserve

Not every edge deserves equal weight. Identify behavior-relevant local structure and apply structure-preserving constraints selectively instead of reconstructing the full neighborhood.

04 / Evaluate

Evaluate at fixed capacity

Compare structure retention before and after quantization on real behavior graphs, then test generative recommendation at a fixed codebook budget with geometry and constraint ablations.

A graphlet-level diagnostic

Separates semantic reconstruction from the behavioral topology a Semantic ID actually keeps.

Curvature as a resource

Moves beyond a single global geometry by assigning geometric bias per item and per codebook level.

Utility-aware preservation

Tests whether selective structural constraints buy recommendation quality at equal discrete capacity.

Small steps, one claim at a time

Phase 01 Current

Diagnose

Quantify graphlet retention across RQ-VAE checkpoints and domains; correlate retention with downstream performance.

Phase 02 Next

Model

Prototype mixed-curvature codebooks and per-item, per-level allocation; track codebook utilization and stability.

Phase 03 Next

Constrain

Add selective structure-preserving objectives; ablate graphlet types and constraint strength.

Phase 04 Later

Evaluate

Fixed-budget comparisons on real behavior graphs, multiple backbones, and a written analysis.

Open Q1

How do we define utility-relevant topology without leaking downstream labels into training?

Open Q2

Can mixed-curvature codebooks remain well utilized when geometry is reallocated across items?

Open Q3

Does graphlet retention transfer across domains and generative recommender backbones?

Working research proposal. The observation is preliminary and directional; no final benchmark result is claimed on this page. The sequence is designed so that each phase can falsify the next.

Contact

A problem worth a conversation

If you work on generative recommendation, hyperbolic representation learning, or discrete item representations — or if you see a failure mode in this plan — I would like to hear it.