- Rating
- 1
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper proposes STIR, a framework for controlling sampling stochasticity in generative sequential recommendation. For a fixed training interpolant, it fits velocity and score fields and uses them in an SDE whose stochasticity level can be selected at inference from the user's history. With exact fields and appropriate regularity, the score correction compensates for the injected noise and preserves the conditional interpolant marginals. Setting epsilon to zero yields an ODE, which is deterministic conditional on its initial noise draw.
The paper relates specific interpolant and sampler configurations to flow matching, Schrodinger bridges, DDPM, and DDIM; the empirical reconstruction study uses three interpolant-specific checkpoints. It further analyzes path choice and user-dependent stochasticity using a Gaussian-mixture model of the target. The practical personalization scheme clusters history representations into five groups and selects a stochasticity level for each group on validation data. Experiments on seven public benchmarks examine recommendation accuracy, sampling dispersion, catalog coverage under accuracy constraints, integration with DreamRec and PreferDiff, and prediction of multiple future items.
Critical strengths
-
Treating the sampling procedure as a configurable component of a generative recommender is practically useful. The separation of velocity and score estimation provides a convenient way to explore stochasticity choices after training a fixed interpolant. The conditional marginal-preservation argument explains why the score correction is necessary and why conditioning the level on a fixed user history differs from making it depend on the evolving sample state.
-
The empirical investigation is broader than a single accuracy comparison, it includes seven public benchmarks, multiple seeds, post-hoc diversification baselines, synthetic analyses, and two existing recommendation backbones. The group-based policy also improves the reported multi-target NDCG in all six settings of Table 3.
Critical weaknesses
-
The novelty argument and baseline comparison need stronger support. DDIM already allows inference-time stochasticity control without retraining, and selecting its noise level based on a fixed user history does not inherently require changing the training schedule. A comparison with existing samplers using the same user-conditioned tuning would help clarify the benefit of STIR's learned fields and interpolant beyond sampler tuning itself. The unification claim also needs clarification, Section 5.1 uses three checkpoints, so the experiment does not demonstrate exact recovery of all four processes from a single fixed checkpoint.
-
Some theoretical conclusions go beyond what the proofs establish. The uniqueness claim in Theorem 5(a) does not hold, since a quadratic time schedule can also give finite irreducible velocity loss. The history-length argument omits a nonzero within-mode loss, so it does not establish that the loss vanishes or justify the proposed switch-off rule. Lemma 12 only addresses Top-K set membership, not the ranking within the set, and therefore does not establish invariance of NDCG or MRR or the claimed limitation of re-ranking. These gaps weaken the theoretical support for the recommendation-specific conclusions.
Questions for authors
-
Can the authors compare STIR against a fixed diffusion checkpoint using a user- or group-conditioned DDIM within its admissible range, with matched history groups, validation budgets, sampling budgets, and evaluation protocols? This would help separate the benefit of personalized sampler selection from that of STIR's learned fields and interpolant.
-
Could the authors address the concerns above about Theorem 5(a), Proposition 7(b), and Lemma 12, and clarify the scope of these claims? Please also point to or provide the proof of the optimal noise envelope in Theorem 5(b); Appendix H.1 proves Theorem 3, while H.2 covers only 5(a).
-
Eqs. (12)-(14) imply a per-sample score-regression target of , whereas Eq. (15), as written, uses . Is this a typographical error, and which sign was used in the implementation? Please also clarify the scope of the claimed equivalence to native training objectives after adding the retrieval loss.
-
What does "recovery" mean in Section 5.1: equivalence of sampling equations, equality of population minimizers, or agreement of empirical results? On ML-1M, Table 2 reports NDCG@20 of for native DDPM and for the corresponding VP-STIR configuration. How does this difference relate to the recovery claim, and what criterion is used to judge agreement?
-
How sensitive is the group-based rule to the number of clusters and the validation sample size? For the configurations selected under each accuracy budget, please report test NDCG@20 alongside coverage, together with variability across validation subsets and results for short-history users.
2 comments
Thank you for your response. My core concern about novelty remains unresolved, and I will maintain my score.
Reviewer PqHW
We thank the reviewer for the careful technical review.
W1, Q1: DDIM η baseline
We ran this ablation on a frozen VP checkpoint, with STIR's groups (C=5), 512 validation users and protocol.
Per-cluster η loses 4–6% NDCG@20 against the best global η; its +0.02–0.03 coverage only moves along the same trade-off. On VP, raising η even reduces coverage (0.746 → 0.690), whereas STIR's ε increases it (Fig. 2a, 3).
Why η cannot be personalized. By Thm. 6, the optimal level depends only on r(h)=δb(h)/δs(h) (Eq. 10), which users' modes and history move (Prop. 7). On VP, b=ωvp(x+s) (App. D.3): both fields come from one noise head, so δb and δs are tied, r(h) barely varies, and all groups get the same level. STIR fits two separate heads (Eq. 7), so ϵ⋆(h) can vary with the user.
W1, Q4: Meaning of recovery
Since the path is fixed at training, recovery is meant per path: one checkpoint recovers every process defined on its path: (I_lin, γ_σ) gives FM (ε=0) and SB (ε=σ²/2), (I_vp, 0) gives DDIM and DDPM; (I_lin, 0) is only an envelope ablation. It holds at three levels: same SDE (all four), same population minimizer (exact for FM, DDPM; at the optimum for SB, DDIM; App. D). The ML-1M DDPM gap comes from the schedule, not from ε. Native DDPM's discrete schedule (T=50, β up to 0.02) gives ᾱ_T ≈ 0.60 (x_T ≈ 0.78 x₁ + 0.63 x₀): it never trains on pure noise yet samples from it. VP-STIR's continuous schedule gives ᾱ(1) ≈ 5e-5, so training matches sampling (0.0637 vs 0.0442). We clarify this in §5.1.
W2, Q2: Theorem 5(a)
The uniqueness in Theorem 5(a) refers to the transport cost: the linear path is the only affine path minimizing it. The finiteness of the irreducible velocity loss was stated in contrast to the variance-preserving path, on which it diverges (App. H.2), not as a second uniqueness claim; we agree the wording suggested otherwise. The quadratic schedule α=1−t2, β=t2 is indeed our straight line with a different clock, xt=(1−s)x0+sx1, s=t2, with finite loss ∫01s˙V(s)ds≤2∫01V(s)ds. It is excluded by the transport cost: ∫01E∥x˙t∥2dt=∫01α˙2dtE∥x0∥2+∫01β˙2dtE∥x1∥2 ≥ E∥x0∥2+E∥x1∥2, by Cauchy–Schwarz, with equality iff α=1−t, β=t; the quadratic clock costs 4/3 per term. We rephrase 5(a): the linear path uniquely minimizes the transport cost, and its irreducible velocity loss is finite, whereas it diverges on the VP path.
Q2: Proof of Theorem 5(b)
It was mistakenly left out and is now App. H.3.
W2, Q2: Proposition 7(b)
We agree. By App. H.2, Lb∗(h)=∫01Qdt+∫01Υ2E[Varm∣xt(μm)]dt. History removes only the mode-confusion term, at rate O(e−cT); the within-mode term ∫01Qdt>0 remains. We restate 7(b): ϵ⋆(h) decreases with T toward the level this term sets.
W2, Q2: Lemma 12
Indeed, ΔK fixes the Top-K set, not its order. As standard (SASRec, DreamRec), we evaluate on one held-out item (App. B), so NDCG and MRR depend only on its rank r. We add: if ∥x^′−x^∥<Δr/2, with Δr the smaller score gap between the target and its two neighbours, r is unchanged, hence NDCG and MRR. HR@K, used in Fig. 5, needs only ΔK.
Q3: Sign in Eq. (15)
A typo, now fixed: the target is −ξt/ς(t). The code samples with s^=−e^/ς(t), so results are unaffected. The equivalence covers the regression part; natives carry the same retrieval term (App. C).
Q5: Sensitivity of the group rule
We ran all four analyses (new appendix). ML-1M, τ=1.00: each group gets the ε with the highest coverage keeping validation NDCG@20 at the deterministic level, chosen on 512 validation users. Values are relative to C=5 (Cov@10 67.6%, NDCG@20 0.0814); ± is the SD over 5 validation subsets.
Larger C buys coverage at the cost of test NDCG, as each choice rests on fewer users (§5.3): C=20 gains 21% coverage for 9% NDCG, C=100 only 28% for 37%. More validation users make the choice more reliable:
We recommend C ≤ 5 and ≥1,024 validation users; all budgets are in the appendix.