- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
This paper proposes DAME, a graph recommendation framework that combines structural mixture-of-experts with depth-aware rotary encoding. It addresses two challenges: a shared graph propagation structure may mix diverse user preferences and noisy interactions, and directly pooling layer representations may obscure the distinction between shallow and high-order collaborative signals. Experiments on Gowalla, Yelp, and Amazon compare DAME with 19 baselines. The authors report improvements across all evaluated metrics, with the largest gains on Amazon.
Critical strengths
- The motivation of this paper is clear and easy to understand.
- The improvement of this methods is pretty significant.
- Ablation study is implemented to show the effectiveness of the methods.
Critical weaknesses
- it seems that the improvement on amazon dataset is very significant but on other datasets, the results seems to be modest. Is it possible to explain the differences? Is it related to dataset character?
- This method is only tested on 3 datasets. The improvement is less convincing. Is it possible to test it on other datasets?
- seems to be a very important hyperparameter but it is not tested in sensitivity analysis. Can you explain how is selected?
- Using RoPE to encode propagation depth and improve cross-layer fusion is reasonable. However, I remain unconvinced that it preserves differences between nodes and alleviates oversmoothing. Could the authors clarify this mechanism and provide more direct evidence?
- Although the appendix provides a theoretical complexity analysis, the paper lacks empirical comparisons of training time, inference time, total parameter counts, and peak GPU memory usage. Could the authors evaluate the method on a substantially larger dataset and compare these costs with representative baselines? Reporting how runtime and memory consumption grow with the interaction graph size would further strengthen the scalability evaluation.
Questions for authors
See Weaknesses
5 comments
W2: Evaluation Across Three Benchmark Datasets
Key Takeaway: Although evaluated on three public benchmarks, Gowalla (location check-ins), Yelp (local business reviews), and Amazon (large-scale e-commerce) deliberately span three distinct application domains, a range in training interactions (), a range in catalog size (), and sparse-to-dense local connectivity ( avg. degree)—further reinforced by 5-seed statistical validation and a controlled graph-scaling sweep.
- Consistent cross-regime superiority: DAME outperforms 19 competitive baselines on every reported metric across all three domains under identical embedding dimensionality, exhibiting regime-consistent module behavior (W1).
- Multi-seed statistical stability (Table 6): Across 5 independent random seeds on all three datasets, relative standard deviations stay tightly bounded within for Recall and for NDCG:
| Dataset | Recall relative std. (5 seeds) | NDCG relative std. (5 seeds) |
|---|---|---|
| Gowalla | 0.56% | 0.63% |
| Yelp | 0.74% | 0.72% |
| Amazon | 1.08% | 1.51% |
- Controlled edge-scale sweep: Evaluating nested interaction subgraphs on Amazon with fixed node sets (W5) confirms predictable sub-linear runtime and memory growth. In the revision, we will explicitly state these domain/scale characteristics and discuss how catalog heterogeneity and local density govern expected gains.
W3: Selection Rationale and Sensitivity of Mask Threshold
Key Takeaway: In Eq. (6), is the hard binarization cutoff in the straight-through estimator (not a Gumbel-Softmax temperature). Setting corresponds algebraically to the exact zero-logit sign boundary (), requires zero per-dataset tuning, and exhibits minimal variation () across perturbations.
Specifically, Eq. (6) generates discrete edge masks via:
Because , the condition is strictly equivalent to (positive logits retain an edge; negative logits prune it), providing a natural parameter-sign boundary shared across all datasets without tuning. Sweeping on Gowalla under held-out validation confirms strong stability:
| Threshold | Relative Recall@20 vs. | Relative NDCG@20 vs. |
|---|---|---|
| 0.475 | -0.56% | -0.55% |
| 0.490 | -0.24% | -0.18% |
| 0.500 (default) | 0.00% | 0.00% |
| 0.510 | +0.04% | 0.00% |
| 0.525 | -0.04% | -0.06% |
Across the entire interval, Recall@20 and NDCG@20 vary within and , with tying for best NDCG@20 and within of peak Recall@20 while keeping all layer-experts active (Reviewer wAt9, W1). We will clarify 's exact role next to Eq. (6) and add this sensitivity table in the revision.
W4: Does RoPE Preserve Node Distinctions and Mitigate Over-Smoothing?
Key Takeaway: Yes. RoPE strictly preserves within-layer node geometry via orthogonal transformations while disambiguating shallow vs. high-order collaborative signals during cross-layer summation via relative-depth inner products—outperforming learned depth-encoding alternatives with zero extra parameters and cutting 2-to-8 layer Recall degradation by more than half.
-
Exact within-layer geometry preservation (Isometry): At depth , multiplying all nodes by the block-diagonal orthogonal rotation () leaves pairwise cosines, Euclidean distances, and norms strictly invariant:
\left\|R_l h_u^{(l)} - R_l h_v^{(l)}\right\|_2 = \left\|h_u^{(l)} - h_v^{(l)}\right\|_2.$$ Empirically, pre-/post-rotation numerical error is $<4.8\times 10^{-7}$ (FP32 precision), and pre-rotation layer-4 within-layer user/item mean cosines remain low at $0.032 / 0.025$ (far from collapse $\to 1$). Thus, RoPE never distorts within-hop node distinctions. -
Relative-depth cross-layer pooling (): Unlike uniform or scalar-weighted summation—which conflates shallow and high-order hops in one shared subspace—cross-layer inner products under RoPE satisfy , injecting explicit relative-hop awareness with zero extra parameters and unchanged embedding dimensionality.
-
Controlled depth-encoding ablation and 2-to-8 layer scaling (Gowalla): Under identical held-out validation controls (Reviewer pHsz, W2), parameter-free RoPE improves Recall@20 / NDCG@20 by over no depth encoding, outperforming both learnable additive depth embeddings (, which perturb vector norms/cosines) and learnable scalar layer weights (). Furthermore, across 2 to 8 layers:
| Propagation layers | DAME Recall@20 gain over same-depth pure GCN |
|---|---|
| 2 layers | +0.40% |
| 4 layers | +1.41% |
| 6 layers | +1.21% |
| 8 layers | +1.13% |
From 2 to 8 layers, pure GCN degrades by in Recall@20 whereas DAME degrades by only (less than half), maintaining a clear advantage at every depth.
Response to Reviewer iA5w
We sincerely thank the reviewer for recognizing DAME's clear motivation, strong empirical improvements, and comprehensive ablation design. Below, we address the cross-dataset gain differences (W1), benchmark coverage (W2), binarization threshold τ (W3), RoPE's geometric and anti-oversmoothing mechanism (W4), and computational efficiency and scalability (W5).
W1: Why Is DAME's Performance Gain Much Larger on Amazon?
Key Takeaway: Under the exact same embedding dimensionality as the corresponding baselines on each dataset, node-conditional structural routing yields its largest gains when two structural conditions hold simultaneously: (1) high structural heterogeneity across a large candidate item space, so different nodes genuinely require distinct propagation subgraphs; and (2) sufficiently rich local neighborhoods, providing both multi-hop paths to reweight per node and abundant edge-level supervision to train 32 structural experts per layer.
Comparing these graph statistics alongside our module ablations and routing diagnostics reveals a clear, self-consistent mechanism across the three regimes:
Direct validation from ablations and mask diagnostics: In Table 3, removing Structural MoE on Amazon drops Recall by 0.59%—exceeding the 0.39% drop from removing RoPE—whereas on Gowalla and Yelp, removing RoPE causes larger drops (2.18% and 0.59%) than removing Structural MoE (1.36% and 0.22%). Moreover, all 128/128 layer-experts remain active after training, with a 2.73× higher hard mask keep rate on edges incident to nodes selecting an expert (Reviewer wAt9, W1). We will incorporate this structural regime analysis into the revised manuscript.