- Rating
- 2
- Confidence
- 2
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 2
Read review
Paper summary
The paper proposes VLoRA, a parameter-efficient two-stage framework for adapting vision-language models to multimodal sequential recommendation. It first injects collaborative filtering signals through autoregressive next-item prediction (CAP), then refines item representations with bidirectional attention, masked-token prediction, contrastive learning, and representation preservation (RTF), while training only LoRA adapters. The resulting item embeddings can be precomputed offline and reused by different recommenders. Experiments on six datasets, two sequential recommenders, and multiple VLM backbones show strong in-domain and cross-domain performance.
Critical strengths
- The experimental evaluation is comprehensive, and the main empirical results are convincing. The paper evaluates VLoRA across six datasets, two downstream sequential recommenders, and four ranking metrics, covering both in-domain and out-of-domain settings. VLoRA ranks first in 41 out of 48 settings and second in the remaining seven.
- The method shows promising cross-domain generalization and backbone portability. The consistent improvement suggests that the proposed adaptation recipe is not tightly coupled to a particular dataset or VLM backbone.
Critical weaknesses
- The methodological novelty is somewhat incremental. The overall framework largely combines several established ideas, including autoregressive collaborative alignment, bidirectional embedding refinement, LoRA adaptation, masked-token prediction, contrastive learning, and representation regularization. While their integration for multimodal recommendation is reasonable and empirically effective, it is less clear which component constitutes a sufficiently novel methodological contribution on its own.
- The evaluation scenarios are still relatively limited. The experiments consider only one title and one image per item, interaction sequences of length at most 10. In addition, the learned item representations are static and do not depend on the user or sequence position. Therefore, it remains unclear how well the method scales to richer multimodal metadata, longer behavioral histories, or more modern and large-scale recommendation architectures.
Questions for authors
- Which part of VLoRA should be regarded as the main methodological novelty? Many individual components—two-stage adaptation, LoRA, bidirectional attention, contrastive learning, and representation regularization—have appeared in prior work. Could the authors clarify which design choice they consider the central novel contribution beyond their combination?
- How does VLoRA behave with richer item content and longer user histories? The current setup uses one title, one image, and sequences of length at most 10. It would be helpful to understand whether the gains remain when items contain multiple images, longer textual descriptions, or when user histories are substantially longer.
- The appendix shows that 4→4 slightly outperforms 4→8 on Arts, while 4→8 performs better on Goodreads and on average. Could the authors provide more intuition or analysis explaining when progressive layer expansion is expected to help?
4 comments
Response to Reviewer aeLt (2 / 3)
2. W2 & Q2: Scaling to Longer User Histories () and Richer Item Content
Core Conclusion: VLoRA's decoupled item-level architecture scales to longer user histories () and richer item metadata (adding categories and descriptions) with consistent accuracy gains and zero additional VLM memory or compute cost (Appendix I.5, Tables 20–21 of the revised manuscript).
(a) Longer User Interaction Histories ()
In our main experiments, was used because the average sequence length across the 5-core Amazon benchmarks is interactions per user (Table 2). Crucially, when scaling to longer histories ( or ), producing static, context-independent item embeddings is a key architectural advantage:
- In sequence-encoder VLMs [6] (which feed the entire multi-image history into the VLM), increasing from to expands the input to visual tokens, causing attention blow-up and GPU out-of-memory (OOM) failure.
- In VLoRA, every item is encoded offline once into and sequence dynamics are modeled by the lightweight downstream recommender (
SASRec[5] /GRU4Rec). Extending the history window from to or therefore incurs zero extra VLM memory or inference cost.
We evaluated downstream SASRec [5] with on Goodreads (which has the longest user histories in our benchmark, averaging interactions per user):
Table R2. Scaling to Longer User Interaction Histories on Goodreads (SASRec; Revised Table 20)
| Downstream History Window | Sequence-Level VLM Token Load | VLoRA Additional VLM Cost | LLM2Rec (Qwen3-1.7B) R@20 | VLoRA R@20 | VLoRA N@20 | Relative Gain over LLM2Rec |
|---|---|---|---|---|---|---|
| (Main paper setting) | ~2.5k–3k tokens / seq | Baseline (offline once) | 0.2141 | 0.2527 | 0.1425 | +18.0% |
| self-attention FLOPs | 0 extra VLM calls / memory | 0.2208 | 0.2598 | 0.1468 | +17.7% | |
| OOM ( visual tokens at ) | 0 extra VLM calls / memory | 0.2221 | 0.2614 | 0.1479 | +17.7% |
As shown in Table R2, VLoRA's performance improves steadily as increases to (0.2598) and (0.2614), maintaining a +17.7% lead over LLM2Rec [1] with zero additional VLM overhead.
(b) Richer Item Content (Adding Category and Product Descriptions)
To test whether VLoRA's gains hold when richer textual metadata is supplied, we evaluated VLoRA and same-backbone LLM2Rec (w/o RPR) on Arts using Title + Category + Product Description alongside the product image (Appendix I.5, Table 21):
Table R3. Performance with Richer Item Metadata on Arts (SASRec; Revised Table 21)
| Item Metadata Supplied to Encoder | Same-Backbone LLM2Rec (w/o RPR) R@20 | Same-Backbone LLM2Rec (w/o RPR) N@20 | VLoRA (Multimodal) R@20 | VLoRA (Multimodal) N@20 | Multimodal Gain (R@20) |
|---|---|---|---|---|---|
| Title + 1st Image (Main benchmark) | 0.1404 | 0.0781 | 0.1559 | 0.0875 | +11.0% |
| Title + Category + Description + 1st Image | 0.1468 | 0.0818 | 0.1604 | 0.0902 | +9.3% |
Adding category and description text improves both encoders (0.1559 -> 0.1604 for VLoRA), and VLoRA maintains a +9.3% multimodal gain because visual appearance (color, pattern, craft style) provides complementary information not captured by text descriptions. Both scaling studies have been added to Appendix I.5 (Tables 20–21) of the revised manuscript, with updated cross-references in Section 5 and Appendix C.3.
Response to Reviewer aeLt (3 / 3)
3. Q3: When Progressive Layer Expansion (4 -> 8 vs. 4 -> 4) Helps
Core Conclusion: Consistent with surgical layer-wise adaptation theory under distribution shifts [4], across all five schedules in Appendix G.3 (Table 9), restricting Stage 1 to the top 4 layers (4 -> *) is consistently superior to wider Stage-1 adaptation (8 -> *, 28 -> 28), while expanding from 4 to 8 layers in Stage 2 (4 -> 8) specifically improves cross-platform out-of-domain generalization (+2.1% R@20 and +2.6% N@20 on Goodreads) where visual and lexical distributions shift from the source domains.
Schedule (Stage 1 -> Stage 2) | Trainable Params (Stage 1 / Stage 2) | Arts R@20 (In-Domain) | Arts N@20 (In-Domain) | Goodreads R@20 (Cross-Platform OOD) | Goodreads N@20 (Cross-Platform OOD) | 4-Metric Average |
|---|---|---|---|---|---|---|
4 -> 4 (fixed narrow) | 0.46M / 0.46M | 0.1564 | 0.0878 | 0.2476 | 0.1389 | 0.1577 |
4 -> 8 (progressive, default) | 0.46M / 0.92M | 0.1559 (−0.3%) | 0.0875 (−0.3%) | 0.2527 (+2.1%) | 0.1425 (+2.6%) | 0.1597 |
8 -> 8 (fixed wide) | 0.92M / 0.92M | 0.1533 | 0.0855 | 0.2430 | 0.1317 | 0.1534 |
8 -> 4 (reverse narrowing) | 0.92M / 0.46M | 0.1516 | 0.0849 | 0.2424 | 0.1338 | 0.1532 |
28 -> 28 (all -> all) | 3.21M / 3.21M | 0.1531 | 0.0863 | 0.2462 | 0.1367 | 0.1556 |
- Why Stage 1 must remain narrow (
4 -> *beats8 -> *and28 -> 28on every metric) [3, 4]: Stage 1 (CAP) optimizes causal next-title generation over Amazon sequences. Because the loss supervises only text tokens, unfreezing deeper layers in Stage 1 overwrites intermediate cross-modal representations with domain-specific text generation features [3, 4], hurting both in-domainArtsand OODGoodreads. Restricting Stage 1 to the top 4 layers confines collaborative alignment to the top readout layers while halving Stage-1 trainable parameters. - Why
4 -> 8helps on cross-platformGoodreads(+2.1%R@20) while tying on in-domainArts(−0.3%) [4]:- On in-domain
Arts, items come from the same Amazon distribution seen in Stage 1; adapting the top 4 layers in Stage 2 (4 -> 4) is already sufficient to transition from causal to bidirectional pooling (0.1564vs.0.1559, within seed variance). - On cross-platform OOD
Goodreads, book covers and titles differ substantially in layout and style from Amazon product thumbnails. Unlocking layers 5–8 in Stage 2 (4 -> 8) under bidirectional MMNTP and RPR provides the extra upper-middle-layer capacity needed for cross-modal realignment under distribution shift [4], driving the+2.1%R@20 and+2.6%N@20 improvement (clarified in Section 3.2 and Appendix G.3 of the revised manuscript).
- On in-domain
References
- [1] He et al. LLM2Rec: Large Language Models Are Powerful Embedding Models for Sequential Recommendation. KDD 2025.
- [2] BehnamGhader et al. LLM2Vec: Large Language Models Are Secretly Powerful Text Encoders. COLM / arXiv:2404.05961, 2024.
- [3] Kumar et al. Fine-Tuning Can Distort Pretrained Features and Underperform Out-of-Distribution. ICLR 2022.
- [4] Lee et al. Surgical Fine-Tuning Improves Adaptation to Distribution Shifts. ICLR 2023.
- [5] Kang & McAuley. Self-Attentive Sequential Recommendation (SASRec). ICDM 2018.
- [6] Kim et al. VLM2Rec: Resolving Modality Collapse in Vision-Language Model Embedders for Multimodal Sequential Recommendation. arXiv:2603.17450, 2026.
Dear Reviewer,
Thank you again for your careful review and thoughtful feedback. As the discussion period draws to a close, we would like to briefly follow up on our responses and the corresponding manuscript revisions.
When you have a chance, we would be grateful to hear whether these have helped address your concerns, or whether any points would benefit from further clarification. We would be happy to provide any additional explanation.
Thank you for your time and consideration.
Best regards, The Authors
Response to Reviewer aeLt (1 / 3)
We thank Reviewer aeLt for recognizing the comprehensiveness of our evaluation across six datasets and two sequential recommenders (ranking first in 41 of 48 settings and second in 7), as well as VLoRA's cross-domain generalization and backbone portability. Below we address W1–W2 and Q1–Q3 directly with controlled experiments, all of which—along with presentation revisions—have been incorporated into the revised manuscript (highlighted in blue).
1. W1 & Q1: Central Methodological Novelty Beyond Combining Existing Components
Core Contribution: Beyond integrating two-stage adaptation [1, 2] with LoRA, VLoRA's central methodological contribution is identifying and resolving the Pooling–Manifold Coupling that arises when converting a decoder-only VLM into a recommendation item encoder under severe modality-length asymmetry.
Specifically, adapting a VLM introduces a structural bottleneck absent in text-only LLMs (
LLM2Rec[1]):+4.1%R@20 gain when images are added (0.1398 -> 0.1456on Arts, Table R1).1.000to0.563(Table 10) and reducing Arts R@20 from0.1494to0.1423(Table 6).0.708) and modality-asymmetric text masking (which keeps visual patches intact to reconstruct masked title tokens) prevents visual manifold drift and raises Arts R@20 to0.1559(+11.5% over the text-only port, +11.0% over same-backboneLLM2Rec[1]).A 2×2 controlled study on the exact same
Qwen3-VL-2B-Instructbackbone (Appendix G.1, Table 6 of the revised manuscript) confirms that this coupling is specific to multimodal adaptation:Table R1. Isolating the Multimodal Pooling–Manifold Coupling on
Qwen3-VL-2B-Instruct(SASRec; Revised Table 6)Qwen3-VL-2B-InstructBackbone)LLM2Rec[1] (Text-only, Mean pooling)LLM2Rec[1] + RPRIn the text-only setting on
Qwen3-VL-2B-Instruct, RPR changes R@20 by only+1.2%(0.1404 -> 0.1421) on Arts and+0.8%(0.2416 -> 0.2435) on Goodreads. In the multimodal setting, RPR improves content mean pooling by+9.6%on Arts (0.1423 -> 0.1559) and+5.2%on Goodreads (0.2403 -> 0.2527).Manuscript Revisions for Presentation Clarity (Abstract, Section 1, Section 3.2, Section 4.3, and Appendix G.1): In the revised manuscript (highlighted in blue), we have expanded Appendix G.1 (Table 6) from a 1D cumulative chain into the 2×2 orthogonal control (Table R1) and revised the Abstract, Introduction, Section 3.2, and Section 4.3 to foreground the pooling–manifold coupling and its scope conditions upfront.