- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes SPRINT, which applies MeanFlow’s average-velocity perspective to discrete semantic ID (SID) generation. It derives average token posteriors from the average probability velocity and uses a single Transformer forward pass to predict a complete SID. The model combines token-level contrastive learning with SID-level contrastive matching. The paper evaluates recommendation accuracy and inference efficiency on eight datasets.
Critical strengths
-
The problem is practically relevant: autoregressive decoding and iterative denoising increase inference latency in generative recommendation.
-
The two-level objective addresses position-wise prediction and whole-SID matching; the ablations show that both components contribute.
-
The experiments cover multiple datasets and report recommendation accuracy, inference efficiency, and component ablations.
Critical weaknesses
-
The connection between the central theory and the optimized objective is unclear. The average-probability-velocity derivation leads to a cross-entropy objective over average token posteriors. The final objective replaces this with token-level candidate ranking and adds an SID-level matching term that is not derived from APV. Inference also does not explicitly compute or integrate a velocity. The paper therefore does not establish what APV contributes beyond direct one-step SID prediction. A matched baseline using the same architecture and training objective, but trained as a direct one-step SID predictor, would help isolate this contribution.
-
The sources of the performance gains are not adequately separated. The ablations show that the SID-level branch contributes substantially, but the paper does not add this matching branch to baselines such as RPG or OneGR. Moreover, w/o Contrast removes both negative-sample contrast and SID-level matching, so it cannot measure their individual contributions.
-
The scope of the inference-efficiency claims needs clarification. A single Transformer forward pass does not make catalog search independent of the number of items. The released code scores every catalog item before returning the top K, so computation and cached-item memory grow with catalog size. Given that methods such as RPG use graph-constrained search, the paper should compare graph- or ANN-based candidate search, or clearly scope its efficiency claims to reranking a pre-retrieved candidate set.
-
The released code does not fully match the reported experimental settings. The paper reports \rho=0.75 for Scientific and Instruments, whereas the code fixes the training proportion at 1.0. The code also uses an SID-level softmax temperature that is not reported in the paper’s equations or hyperparameter table. In addition, the paper reports results averaged over five random seeds, but the released configuration and run instructions do not provide a corresponding multi-seed procedure.
-
The negative-sampling implementation may not match the paper’s definition. The paper describes negatives as items the user has not interacted with. The code samples with replacement from the global catalog and excludes only the current target SID; it does not filter the user’s history. Thus, previously interacted items may be used as negatives.
Questions for authors
Please refer to Weaknesses.
9 comments
Responses to Reviewer meP8 (Part 2)
Weakness 2: Separating the sources of the gains: "the paper does not add this matching branch to baselines such as RPG or OneGR. Moreover, w/o Contrast removes both negative-sample contrast and SID-level matching."
Thank you for this suggestion. We add both experiments.
(1). RPG has better performance with our SID-level flow objective.
We have added our SID-level branch to RPG, Table 1 shows the results. From the results, we can have the following finding:
Key Finding: RPG can achieve better performance with our SID-level flow objective.
Table 1. Performance comparision between RPG and RPG + SID-level flow objective.
| Method | Sports | Beauty | Toys | Scientific | Instruments | Games | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | |
| RPG | 0.0463 | 0.0263 | 0.0809 | 0.0464 | 0.0869 | 0.0490 | 0.0395 | 0.0218 | 0.0545 | 0.0300 | 0.0853 | 0.0485 |
| RPG + SID-level flow objective | 0.0503 | 0.0279 | 0.0865 | 0.0490 | 0.0918 | 0.0494 | 0.0454 | 0.0245 | 0.0604 | 0.0326 | 0.0920 | 0.0504 |
| Improv. | 8.69% | 6.25% | 6.91% | 5.62% | 5.59% | 0.88% | 14.94% | 12.39% | 10.83% | 8.67% | 7.85% | 3.92% |
(2). Contrastive formulation is necessary and contribute positively to our training.
To address your concern, I have design three variants :
- V1: replacing the token-level contrastive objective with ordinary per-token cross-entropy while keeping others unchanged.
- V2: replacing the SID-level contrastive objective with ordinary cross-entropy while keeping others unchanged.
- V3: replacing both token-level and SID-level contrastive objective with cross-entropy.
Table 2 reports the performance of our method and three variants. We have the following finding.
Key Finding: Removing either side contrastive objective harms model performance to a great extent.
Table 2. Performance comparison between different variants in terms of contrastive formulation.
| Method | Sports | Beauty | Toys | Scientific | Instruments | Games | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | R@10 | N@10 | |
| SPRINT (ours) | 0.0621 | 0.0347 | 0.0987 | 0.0552 | 0.0998 | 0.0594 | 0.0656 | 0.0369 | 0.0713 | 0.0386 | 0.1001 | 0.0534 |
| V1 | 0.0466 | 0.0247 | 0.0821 | 0.0463 | 0.0900 | 0.0547 | 0.0482 | 0.0280 | 0.0515 | 0.0280 | 0.0745 | 0.0408 |
| V2 | 0.0442 | 0.0234 | 0.0831 | 0.0460 | 0.0886 | 0.0544 | 0.0432 | 0.0248 | 0.0474 | 0.0259 | 0.0696 | 0.0374 |
| V3 | 0.0283 | 0.0155 | 0.0645 | 0.0381 | 0.0756 | 0.0477 | 0.0307 | 0.0187 | 0.0324 | 0.0188 | 0.0537 | 0.0310 |
Responses to Reviewer meP8 (Part 3)
Weakness 3: Scope of the inference-efficiency claims: "A single Transformer forward pass does not make catalog search independent of the number of items ... compare graph- or ANN-based candidate search, or clearly scope its efficiency claims."
Thank you for this comment. We have provided detailed complexity analysis of graph-constrained search of RPG and ANN-based candidate search of our method SPRINT below.
RPG. RPG's search and scoring consist of two parts. The first is the construction of its decoding graph. For each of the valid SIDs, RPG retrieves nearest neighbours, and each similarity costs , where is the SID length. This gives time and memory. The graph must also be maintained as the catalog changes. Each new item requires time to find its neighbours, and the adjacency lists of those neighbours must then be updated. The second part is online decoding. Starting from a beam of SIDs, RPG runs iterations. In each iteration, it expands the neighbours of every beam SID, scores each candidate with additions, and keeps the top . This costs per user. The overall complexity is therefore . The iterations must run sequentially, and each one involves irregular memory access over the graph.
SPRINT. SPRINT requires no decoding graph. After one forward pass, it scores all valid SIDs directly. The token-level term sums table lookups per SID, which costs . The SID-level term is a single matrix product between and the precomputed , which costs . The overall online complexity is therefore . All of these operations are dense and fully parallel, and they finish in one round. The SID representations are computed once offline in . A new item only needs one MLP pass to compute its , at a cost of , and its representation is then appended, with no graph to rebuild or update.
Comparison.
(1) Offline preparation.
RPG builds its decoding graph in , while SPRINT precomputes in . RPG is therefore more expensive whenever , which holds for typical catalog sizes, and the gap grows linearly with .
(2) Catalog updates.
For each new item, RPG needs to find its neighbours, while SPRINT needs to compute its . RPG is therefore more expensive whenever , and its cost keeps growing with the catalog, whereas SPRINT's cost stays constant.
(3) Online scoring per user.
RPG’s online decoding costs , independent of , but it relies on a decoding graph that costs to build. SPRINT needs no graph. It scores all SIDs in with a single, fully parallel matrix product, whereas RPG requires sequential rounds of irregular graph access. This is why SPRINT is substantially faster in practice (Table 3). Moreover, as stated in our manuscript (refer to Appendix F), SPRINT is more suitalbe in the ranking stage, where a candidate set of size has already been retrieved. There, SPRINT costs only , whereas RPG would have to rebuild its graph for every candidate set at . SPRINT therefore has a lower complexity than RPG in the ranking stage.
Responses to Reviewer meP8 (Part 1)
Thank you for your thorough review and for carefully examining our released code. We appreciate your recognition of the practical relevance of the problem, the complementary roles of the two objective levels, and the breadth of our experiments. Below are our detailed responses. We would be glad to discuss further.
Thank you for this comment. Our objective is derived from the average probability velocity (APV), and we have added two formal results in the updated version to make this explicit (refer to Appendix C.5 and C.6 in Page 23-25 in our updated version).
(1). The token-level objective is APV matching.
Eq. (8) aligns the marginal and conditional APVs (refe to Appendix C.5). Subtracting the marginal APV (Eq. 6) from the conditional one (Eq. 5), the current-state terms cancel and the shared scalar factors out:
vˉr,tl(xl,z∣x1)−vˉr,tl(xl,z)=λr,t[δx1(xl)−pˉ1∣r,t(xl∣z)].
The velocity residual thus reduces to a posterior residual. Any consistent velocity-matching loss on this residual and the per-token cross-entropy in Eq. (8) therefore share the same minimizer. Eq. (8) is a valid surrogate for APV matching, not a replacement of it.
Eq. (9) strictly extends Eq. (8) . Let Φ(c)=∑llogpˉ1∣r,t(cl∣xr,H,θ). Then
∇θLCL=∇θLCE+∑c∈Sw(c)∇θΦ(c),w(c)=∑c′∈Sexp(Φ(c′))exp(Φ(c)).
The contrastive gradient contains the full APV-matching gradient, plus a term that suppresses competing SIDs. The contrastive form therefore builds on the APV objective rather than replacing it.
(2). The SID-level term is motivated by the structure of the one-step transition.
The exact one-step transition from the all-mask state is the joint posterior over the whole SID, while a single parallel pass recovers only its per-position marginals. The factorized score therefore cannot express how the tokens of an SID constrain one another. The SID-level term scores the L tokens jointly as one item and supplies exactly this missing dependence. We have explicitly explain this in our manuscript (refer to Sec. 3.4 Lines 310-315 in Page 6)
(3). Inference is the closed-form integration of the APV.
Because the APV has a fixed direction and a closed-form rate (our Proposition 2 in Appendix C.3), it can be integrated exactly. The per-position transition is Pt∣r=δz+(1−e−(t−r)λr,t)(pˉ1∣r,t−δz), which reduces to pˉ1∣0,1 on [0,1]. Reading pˉ1∣0,1 in Eq. (11) is therefore the integrated APV, and no numerical integration is needed.