- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 1
- Soundness
- 2
Read review
Paper summary
The paper proposes OxygenFlowRec, a one-step continuous recommender that retains semantic-ID structure without autoregressively or diffusively generating SID tokens. It predicts a continuous query in one flow-style step, retrieves items using full-catalog cosine similarity, and returns their fixed SIDs through a lookup. Experiments on Amazon datasets compare accuracy and efficiency with dense, generative, and diffusion-based recommenders.
Critical strengths
- The method avoids autoregressive SID decoding and beam search, providing a potentially useful accuracy-efficiency tradeoff.
- The paper reports several ablations, matched controls, collision analyses, and latency measurements.
Critical weaknesses
- The presentation is very difficult to follow. The paper lacks a clear narrative and often introduces symbols and design choices without adequate definitions or motivation. This is especially prominent in section 3. These are central parts of the method, making the paper difficult to understand and reproduce.
- The contribution is insufficiently distinguished from dense retrieval. Inference ultimately produces one continuous query, performs full-catalog cosine retrieval, and looks up the selected items’ fixed SIDs. The brief related-work discussion and limited comparison with the closest methods make it unclear what is fundamentally new beyond a complicated representation and training recipe for dense retrieval.
Questions for authors
- In what precise sense is OxygenFlowRec a generative recommender rather than a dense retriever with SID-structured item embeddings and a fixed SID lookup?
- Can the authors provide clearer, matched comparisons with a capacity-matched dense retriever?
- Can the authors provide a self-contained definition of every quantity in Section 3?
- What is the motivation for frequency-gated residuals, and why should training frequency determine whether an item receives item-specific capacity? What alternatives were considered?
5 comments
The trainable text projection is included in the recommender parameter count. The frozen Sentence-T5 encoder is excluded. Storing one 128-dimensional FP32 text vector per item requires 12.6 MiB on Scientific and 79.1 MiB on Toys, reported separately from model parameters and peak inference memory.
| Model | Scientific total parameters | Scientific R@20 | Toys total parameters | Toys R@20 |
|---|---|---|---|---|
| Direct CE, item-only | 41.06M | 0.0694 | 110.79M | 0.0530 |
| Direct CE, near-budget-matched dense | 41.16M | 0.0700 | 110.84M | 0.0535 |
| Direct CE, text-enhanced dense | 41.16M | 0.0718 | 110.84M | 0.0555 |
| Direct CE, SID + residual | 41.19M | 0.0713 | 110.92M | 0.0549 |
| Full OxygenFlowRec | 41.19M | 0.0739 | 110.92M | 0.0566 |
The alternatives use the same interaction split, history limit, training budget, exact-item evaluator, and validation selection rule. They are near parameter-budget-matched, not architecturally identical.
Relative to the near-budget-matched dense predictor, SID + residual improves R@20 under Direct CE by 1.9% on Scientific and 2.6% on Toys. Full improves R@20 by 5.6% and 5.8%, respectively. These comparisons directly test whether the proposed model improves over a dense retriever with a similar recommender parameter budget.
The text-enhanced dense model is a stronger comparison than an item-ID-only table. Its Direct CE results exceed Direct CE with SID + residual on both datasets; Full OxygenFlowRec remains higher. We therefore do not claim that SID + residual is uniformly better than every direct dense representation.
The original SID assignments also matter. In Table 4, shuffling complete SID tuples reduces R@20 from 0.0739 to 0.0694 on Scientific and from 0.0566 to 0.0538 on Toys. The shuffle preserves tuple-use counts and collision sizes, so it tests the assignment structure rather than simply the presence of an additional embedding table.
We also compare close continuous methods on Scientific and Instrument:
| Method | Scientific R@20 | Instrument R@20 |
|---|---|---|
| FAVE | 0.0735 | 0.0950 |
| Hybrid Direct CE | 0.0713 | 0.0925 |
| MIRAGE | 0.0739 | 0.0948 |
| Full OxygenFlowRec | 0.0739 | 0.0954 |
MIRAGE and OxygenFlowRec match at the displayed precision on Scientific. The Instrument comparison favors OxygenFlowRec among these rows. These single-seed point comparisons do not establish universal superiority over continuous retrievers. The original MIRAGE comparisons remain available in Appendix M.3.
3. Self-contained definitions for Section 3
Addresses Weakness 1 and Question 3.
We agree that Section 3 should define each quantity and its role before introducing the complete objective. We will reorganize it around item representation, history encoding, training, optimization, and inference, using the following compact definitions.
3.1 Item representation
Let be the catalogue and . For item :
| Symbol | Definition and role |
|---|---|
| Fixed SID tuple assigned to item | |
| Number of codebook positions | |
| Number of codes per codebook | |
| Trainable embedding table for codebook position | |
| , | Code-embedding and item-vector dimensions |
| Trainable item-specific residual | |
| Item frequency in the training interactions | |
| Frequency-dependent residual gate | |
| Residual scale | |
| , | Item representation and its normalized version |
The item representation is
The gate is
Thus, an item observed once has zero residual contribution, and the gate reaches one at nine training occurrences.
The main configuration uses , , , , and .
3.2 User history, state, and history-encoder query token
| Symbol | Definition and role |
|---|---|
| , | User and their history sequence |
| , | History length and next-item target |
| Normalized target-item state | |
| Sampled continuous source state | |
| Training interpolation time | |
| Current interpolated training state | |
| Bound used for training time and the inference source | |
| Learnable query token appended to the history sequence | |
| History context conditioned on state and time |
The source is sampled as with . Training samples and clips it to , where .
The history encoder is a six-layer, eight-head Transformer. Its state-dependent update to the appended query token is
where combines learned state and time projections, , and .
updates the learnable history-encoder query token . It is distinct from the final continuous retrieval query .
3.3 Recovery and inference endpoint
is the target-recovery network with parameters . At an interpolated state, it predicts
At inference, the context is recomputed at the source state:
One update produces the retrieval query
At inference, is generated deterministically from a history hash; the predictor does not access the target-containing training interpolant.
Response to Reviewer mCKZ
We thank the reviewer for the careful assessment. We particularly appreciate the recognition that avoiding autoregressive SID decoding and beam search provides a potentially useful accuracy–efficiency trade-off, and that the paper includes matched controls, collision analyses, and latency measurements.
Below, we address the two weaknesses and four questions point by point.
1. Continuous prediction versus dense retrieval
Addresses Weakness 2 and Question 1.
At inference, OxygenFlowRec is a dense retriever: it produces one continuous query, ranks catalogue item representations by cosine similarity, and returns the selected items’ fixed SIDs through lookup. The SID lookup does not generate SID tokens, alter the item ranking, or resolve SID collisions.
Our use of “continuous generation” describes how the continuous query is trained and constructed. The network learns to recover target-item states from continuous interpolants. At inference, a single update from a source state produces the query used for catalogue ranking.
The system therefore has three distinguishable stages:
We will use “one-step continuous query prediction with catalogue retrieval” as the precise operational description.
The contribution is not the use of dense retrieval alone. Our modeling question is how to train the predictor and item space when fixed SID assignments determine parameter sharing, item-specific residuals provide additional capacity, and catalogue CE supervises the actual one-step inference endpoint.
Thus, SID assignments fix the parameter-sharing structure, while the item geometry and one-step predictor are jointly trained for the recommendation task. The controlled comparisons below test whether these choices improve ranking beyond a direct dense predictor.
2. Comparison with capacity-matched dense retrievers
Addresses Weakness 2 and Question 2.
We agree that a close dense-retrieval comparison is essential. Table 4 already separates the effects of representation and training recipe:
Our closest capacity-matched control is Hybrid Direct CE: it uses the same trainable retrieval parameterization and inference architecture as Full, while differing in the training objectives. Training-only auxiliary components are accounted for separately.
With SID + residual representations, Direct CE obtains R@20 of 0.0713 on Scientific and 0.0549 on Toys, versus 0.0739 and 0.0566 for Full. This comparison changes multiple training objectives, so the entire gap should not be attributed to recovery alone.
The no-recovery control changes only the recovery coefficient within the full recipe. It obtains 0.0720 on Scientific and 0.0555 on Toys, providing a more targeted comparison of recovery in the tested setting.
Near parameter-budget-matched dense alternatives
The existing item-only and hybrid models have similar but not exactly equal total parameter counts. We add two direct dense predictors:
Near parameter-budget-matched dense: Starting from the item-only Direct CE predictor, we increase the hidden width of the four prediction-head MLP blocks from 2048 to 2060 on Scientific and to 2054 on Toys. This adds approximately 0.10M and 0.05M parameters, respectively, while retaining the 512-dimensional output and the same retrieval architecture.
Near parameter-budget-matched text-enhanced dense: We concatenate each item's title, brand, categories, features, and description, encode them with a frozen Sentence-T5-base encoder followed by PCA to 128 dimensions, and add a trainable 128-to-64-to-512 projection of these features to the item-ID vector before normalization. We adjust the prediction-head width so that the total recommender parameter budget remains close to Full.