--======== Review Reports ========-- The review report from reviewer #1: *1: Is the paper relevant to Bigdata? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [_] 4 (Innovative) [X] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [_] 4 (High) [X] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [_] 4 (Good) [X] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to Bigdata users and practitioners? [_] 3 (Yes) [X] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [X] 2 (High) [_] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [_] 4 (Accept: a regular paper) [X] 3 (Weak Accept: could be a poster or a short paper) [_] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Detailed comments for the authors This paper presents a vector quantized meta-learning approach for transferable sequential recommenders (MetaRec). The MetaRec is an ID-only method, which does not rely on overlapping user or additional modality information. It contains a vector quantization module and a meta transfer module, leveraging multiple source domains to enhance target domain performance. Strengths: * This work is relevant to BigData conference. The presentation of this paper is clear, with well-defined motivations. * The investigated problem of transferable sequential recommendation without overlapping users is of great practical value and may interest a broad range of audiences. * The experimental results seem promising in the target domain. The supporting experiments are comprehensive. Weaknesses: * VQRec [1] and TIGER [2] have investigated vector quantization and semantic IDs to learn semantic representations. The proposed vector quantization module seems to lack essential improvements, which greatly weakens the significance of this work's contribution. In addition, this article does not consider the relevant TIGER as a baseline. * Although the performance of MetaRec on the target domain is encouraging, there is still room for improvement compared to text-based transferable methods. More explanation is needed as to why only ID-based forms are considered in this work, and MetaRec is not designed with text information. This can enhance the assessment of the rationality of the motivation. * The references contain some errors. For example, TIGER [2] has been accepted in NIPS 2023. I suggest the author check and update the citations. Reference: [1] Y. Hou, Z. He, J. McAuley, and W. X. Zhao, “Learning vector quantized item representation for transferable sequential recommenders” in Proceedings of the ACM Web Conference 2023, 2023, pp. 1162–1171. [2] S. Rajput, N. Mehta, A. Singh, R. H. Keshavan, T. Vu, L. Heldt, L. Hong, Y. Tay, V. Q. Tran, J. Samost, et al., “Recommender systems with generative retrieval,” In NIPS 2023. ======================================================== The review report from reviewer #2: *1: Is the paper relevant to Bigdata? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [X] 4 (Innovative) [_] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [_] 4 (High) [X] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [_] 4 (Good) [X] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to Bigdata users and practitioners? [_] 3 (Yes) [X] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [_] 2 (High) [X] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [X] 4 (Accept: a regular paper) [_] 3 (Weak Accept: could be a poster or a short paper) [_] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Detailed comments for the authors Overall evaluation: This paper introduces MetaRec, a novel approach for cross-domain sequential recommendation. It leverages vector quantization to map item features into a shared feature space and employs meta-learning to dynamically adjust knowledge transfer based on source-target domain similarity. This method operates without requiring shared users or items across domains. Experiments demonstrate that MetaRec significantly outperforms existing methods on multiple benchmark datasets. strength: 1.The paper introduces a unique combination of vector quantization and meta-learning, which addresses the challenge of cross-domain recommendation effectively without relying on shared users or items. By leveraging vector quantization, the approach maps item features to a shared feature space, facilitating better alignment and transfer of knowledge across domains. The meta-learning component dynamically adjusts the influence of different source domains based on their similarity to the target domain, optimizing the transfer process and improving recommendation performance. 2.MetaRec is efficient in terms of parameters, as it reshapes existing embeddings instead of introducing new ones, reducing the risk of overfitting and maintaining model simplicity. 3.In terms of article writing, the formulas are complete and equipped with detailed explanations, and the images are concise and easy to understand, which can serve as a good reading aid. ======================================================== The review report from reviewer #3: *1: Is the paper relevant to Bigdata? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [X] 4 (Innovative) [_] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [X] 4 (High) [_] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [X] 5 (Excellent) [_] 4 (Good) [_] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to Bigdata users and practitioners? [X] 3 (Yes) [_] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [X] 2 (High) [_] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [X] 4 (Accept: a regular paper) [_] 3 (Weak Accept: could be a poster or a short paper) [_] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Detailed comments for the authors Considering the limitation of existing methods that leverage shared information or additional modalities to generate next-item recommendations, this paper proposes a vector quantized meta learning for transferable sequential recommenders by ignoring the effect of additional modalities or shared information across domains. Finally, the effectiveness of proposal is verified on benchmark datasets. Strong Points 1. Paper is easy to follow and contributions are clear. 2. The proposal is verified based on benchmark dataset. Weak Points 1. How about the training and reference time? 2. How about the trade-off between two loss in equation 10. 3. Implementation details are missing and source code is not available. 4. Please keep the format of reference consistent. ======================================================== The review report from reviewer #4: *1: Is the paper relevant to Bigdata? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [_] 4 (Innovative) [X] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [_] 4 (High) [_] 3 (Good) [X] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [_] 4 (Good) [X] 3 (Above average) [_] 2 (Below average) [_] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to Bigdata users and practitioners? [X] 3 (Yes) [_] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [X] 2 (High) [_] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [_] 4 (Accept: a regular paper) [_] 3 (Weak Accept: could be a poster or a short paper) [X] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Detailed comments for the authors 1.Summary of the Paper: a)Motivation: Sequential recommendation has advanced, but transferring models across domains is difficult due to heterogeneous user and item groups. Existing methods have limitations regarding shared information or additional modalities. b)Contribution: Proposes MetaRec, a vector quantized meta learning approach for transferable sequential recommendation without requiring additional modalities or shared information across domains. 2.Evaluation: a)Strength: i.The proposed MetaRec approach is innovative as it combines vector quantization and meta learning for transferable sequential recommendation in an ID-only and non-overlapping setting. This addresses a significant gap in existing research where most methods rely on shared information or additional modalities. ii.The detailed description of the vector quantization module, including its mapping and learning processes, and the meta transfer approach with its formulation, inner-and outer-level optimization, and gradient rescaling, shows a well-thought-out and comprehensive methodology. iii.The extensive experiments conducted on multiple source-target dataset selections, using various evaluation metrics and comparing with a wide range of state-of-the-art baselines, provide strong evidence for the effectiveness of the proposed method. The ablation studies to understand the contribution of individual modules further enhance the credibility of the results. iv.The approach is designed to be applicable to universal model architectures and recommendation scenarios, as it does not depend on specific domain characteristics or additional input modalities, thus having high potential for generalization. b)Weakness: i.The author states: "Contribution 1: To the best of our knowledge, we are the first to propose a solution for cross-domain sequential recommendation based on an ID-only setting with disjoint item groups." However, a paper [1] has been found that also considers cross-domain recommendation based on an ID-only setting with disjoint item groups. ii.Contribution 2: The proposed vector quantization maps item embeddings across domains to a well-aligned feature space. Moreover, our meta transfer 'learns to transfer' from multiple sources for improved target domain performance. Nevertheless, a paper [2] considers the same quantization method for transferable sequential recommenders. The difference lies in that paper [2] considers VQ in text embedding while this paper focuses on ID embedding. 3.Suggestions for Improvement: a)The author states that there are M source domains (M > 1). It is recommended that the authors consider the "multi-domain recommendation" field instead of the cross-domain field. The statement is not clear regarding whether the item of z_q is from the source domain or the target domain. b)The author does not describe the embedding table of z_q and its relationship with e, making it difficult for readers to understand what the Vector Quantization is doing.While the equations (4) to (9) are unclear. The authors can refer to paper [3] to better present the meta learning method so that readers can more easily understand. c)It is noted that the baseline model LRURec paper conducts experiments in ML-1M, Beauty, Video, Sports, and Steam. It is suggested that MetaRec should also conduct studies in these datasets. 4.Overall Comments: a)Recommendation: Weak Reject b)Justification: While the paper is strong in its methodology and contributions, addressing the noted weaknesses and suggestions would enhance its clarity and impact, particularly for non-expert readers. [1] Zhu Y, Tang Z, Liu Y, et al. Personalized transfer of user preferences for cross-domain recommendation[C]//Proceedings of the fifteenth ACM international conference on web search and data mining. 2022: 1507-1515. [2] Hou Y, He Z, McAuley J, et al. Learning vector-quantized item representation for transferable sequential recommenders[C]//Proceedings of the ACM Web Conference 2023. 2023: 1162-1171. [3] Pan F, Li S, Ao X, et al. Warm up cold-start advertisements: Improving ctr predictions via learning to learn id embeddings[C]//Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval. 2019: 695-704. ======================================================== The review report from reviewer #5: *1: Is the paper relevant to Bigdata? [_] No [X] Yes *2: How innovative is the paper? [_] 5 (Very innovative) [X] 4 (Innovative) [_] 3 (Marginally) [_] 2 (Not very much) [_] 1 (Not) [_] 0 (Not at all) *3: How would you rate the technical quality of the paper? [_] 5 (Very high) [_] 4 (High) [X] 3 (Good) [_] 2 (Needs improvement) [_] 1 (Low) [_] 0 (Very low) *4: How is the presentation? [_] 5 (Excellent) [_] 4 (Good) [_] 3 (Above average) [_] 2 (Below average) [X] 1 (Fair) [_] 0 (Poor) *5: Is the paper of interest to Bigdata users and practitioners? [_] 3 (Yes) [X] 2 (May be) [_] 1 (No) [_] 0 (Not applicable) *6: What is your confidence in your review of this paper? [_] 2 (High) [X] 1 (Medium) [_] 0 (Low) *7: Overall recommendation [_] 5 (Strong Accept: top quality) [_] 4 (Accept: a regular paper) [_] 3 (Weak Accept: could be a poster or a short paper) [X] 2 (Weak Reject: don't like it, but won't argue to reject it) [_] 1 (Reject: will argue to reject it) [_] 0 (Strong Reject: hopeless) *8: Detailed comments for the authors Strengths: 1. The research topic of this paper is important and interesting. 2. The motivation is very convincing. 3. The experiments use a wide variety of datasets and baselines, making the content rich. Weaknesses: 1. The model explanation is not clear enough. Specifically, the VQ part could benefit from more background information, such as explaining what a codebook is. 2. The two-level optimization is not clearly explained. It seems that both levels directly update \theta, so why is it necessary to separate them into two levels? It is recommended to include an algorithm description to clarify the specific optimization process. 3. Regarding this sentence: 'Additionally, MetaRec can be further enhanced by incorporating auxiliary information, or alternatively, transferring from arbitrary source domains.' I don't quite understand how MetaRec is supposed to incorporate auxiliary information. The source domain MetaRec supports should only be in the form of item sequences, right? For example, how would text information be incorporated? ========================================================