Reviews
Review 1
Summary and contributions	This paper introduces a novel diffusion-based framework for sequential recommendation, termed the Dual Conditional Diffusion Model, which jointly leverages implicit and explicit conditioning throughout the diffusion process to balance contextual representation learning with fine-grained signal guidance. Meanwhile, unlike prior works that rely solely on compressed history vectors (implicit) or direct concatenation (explicit), DCRec integrates both conditioning types in the forward and reverse processes through a tailored Dual Conditional Diffusion Transformer (DCDT). The model employs Conditional Layer Normalization, self-attention, and cross-attention to balance global contextual signals with fine-grained historical dependencies. Extensive experiments on three benchmark datasets (Amazon Beauty, Amazon Toys, and Yelp) demonstrate that DCRec achieves state-of-the-art performance, outperforming strong baselines in accuracy and efficiency. Additional ablation, sensitivity, and inference acceleration analyses further validate the effectiveness and practicality of the proposed approach.
Strong points	1) The paper is the first to unify implicit and explicit conditioning in diffusion-based sequential recommendation, addressing a clear methodological gap and offering a principled design that improves representation of user histories.
2) The paper is well-written and easy to follow. It provides sufficient background, detailed methodology, and clear explanations of the proposed solution.
3) The experimental section is thorough and covers all key claims. The authors present overall performance comparisons, ablation and parameter analyses, exploration tests on implicit vs. explicit effects for long and short sequences, and efficiency evaluations showing improved diffusion performance. Furthermore, the release of source code enhances transparency and ensures reproducibility of the results.
Weak points	1) Although the related work section is comprehensive, it occasionally mixes diffusion-based methods with classical sequential models. A clearer separation or taxonomy would help readers quickly locate the most relevant comparisons.
2) The paper shows significant performance on various datasets. However, a brief self-reflection on limitations (e.g., scalability to extremely long sequences) would make the work appear even more balanced.
3) Figure 2 could be improved with larger fonts and clearer legends.
Detailed Evaluation	This paper proposes a new diffusion framework for sequential recommendation—the Dual Conditional Diffusion Model which dynamically integrates implicit and explicit conditions during the diffusion process, balancing contextual modeling capacity with precise signal guidance. The proposed DCDT architecture is technically sound and well-motivated, and the paper is clearly written with sufficient details for reproducibility, further supported by released source code. The experimental evaluation is comprehensive, covering overall performance, ablations, parameter studies, sequence length analysis, and efficiency, with consistent gains over strong baselines. The work is highly relevant to the community and makes a meaningful contribution to advancing diffusion-based recommendation. Minor issues include some mixing of diffusion-based and classical models in the related work section, lack of a brief discussion on potential limitations such as scalability to very long sequences, and figure readability (e.g., Figure 2). These are small presentation improvements that do not undermine the overall quality. Overall, the paper is novel and significant and I recommend acceptance.
Overall rating
4: Accept (Will argue strongly to accept)
Reviewer's confidence
4: Expert: I have published papers on this topic.
Review 2
Summary and contributions	The paper proposes DCRec (Dual Conditional Diffusion for Sequential Recommendation), a novel diffusion-based model designed to improve sequential recommendation systems (SRS). DCRec combines both implicit and explicit conditioning mechanisms within a diffusion model framework to better model sequential user behavior. The proposed approach utilizes both implicit historical behavior embedding and explicit user-item interaction information to guide the recommendation process. The model employs the Dual Conditional Diffusion Transformer (DCDT) to integrate these two sources of information during the forward and reverse diffusion stages. Extensive experiments on benchmark datasets, such as Amazon's Beauty, Toys, and Yelp, demonstrate that DCRec outperforms several state-of-the-art methods in terms of both accuracy and computational efficiency.
Strong points	1. The dual conditioning approach of DCRec, which integrates both implicit and explicit information, is a novel and effective solution for sequential recommendation tasks.

2. DCRec outperforms state-of-the-art models, including generative models like DreamRec and DiffuRec, across multiple benchmark datasets (Beauty, Toys, and Yelp). The improvement in performance, particularly in metrics like HR@5 and NDCG@10, validates the efficacy of the proposed approach.

3. The paper includes extensive experimental analyses, including ablation studies and the impact of different conditioning strategies. This thorough evaluation provides valuable insights into how various components of DCRec contribute to its success.
Weak points	1. The implicit conditioning, which concatenates historical item embeddings with target items, may risk overfitting to the specific patterns present in the historical data. This could lead to a loss of diversity in recommendations, particularly for users with less diverse historical behaviors.

2. Why does DCRec not benefit from longer diffusion inference steps in Figure 8, unlike DiffuRec?

3. Methods like DreamRec are highly dependent on the embedding dimension size, and it seems that the authors did not provide a clear presentation of the hyperparameters for the comparison methods. This makes it difficult for me to easily trust the experimental results."
Detailed Evaluation	see weaknesses
Overall rating
3: Weak Accept (Probable accept, unless convinced otherwise)
Reviewer's confidence
3: Familiar: I have passing knowledge on this topic.
Review 3
Summary and contributions	The paper deals with Sequential Recommendation and more precisely in diffusion-based sequential recommender systems. The authors propose the framework DCRec for Dual Conditional Diffusion Models for Sequential Recommendation. The specificity of the framework is to integrate implicit and explicit information.
Numerous experiments on three datasets (Beauty, Toys and Yelp) show that the framework outperforms several existing models.
Strong points	The description of the framework is well presented and formalized. The motivation of the proposition is well argued.

Evaluation is well conduced with a rigorous evaluation protocol including 3 datasets and 10 baselines.

The results of the experiments are superior to existing models in terms of NDCG@K and HR@K.
Weak points	No future works are given.
Detailed Evaluation	The proposed framework DCRec is interesting. It answers to practical challenges of Sequential recommendation.

Numerous experiments are conducted and the results are encouraging. DCRec outperforms several baselines.

It is important to give some perspectives of the work in the conclusion.
Overall rating
3: Weak Accept (Probable accept, unless convinced otherwise)
Reviewer's confidence
2: Knowledgeable: I have read papers on this topic.
Review 4
Summary and contributions	This paper argues that the existing diffusion-based SR mainly rely on implicit conditional diffusion, which utilizes the compressed user history. The authors propose to model explicit conditional diffusion for sequential recommendation, and present a novel dual conditional diffusion model to exploit both the implicit and explicit modeling of the user historical behaviors.
Strong points	1. The paper is well written and easy to follow.
2. The technical novelty is sufficient since the proposed DCDT exploits the user behaviors with implicit conditioning for forward process, and explicit conditioning for reverse process, in a unified way.
3. The authors conduct extensive experiments to illustrate the superiority of the proposed solution against existing SOTA. Further model analysis is also sufficient and insightful.
Weak points	1. There are some typos and grammatical errors in paper writing.
2. The whole story of explicit and implicit conditioning is not distinct enough. The authors should deliver a formal definition for the implicit and explicit conditioning respectively. Specifically, according to the description, the citaion [18] could also join the explicit conditioning line.
3. The model conditioning analysis should be further enriched to tell why the explicit component preserves temporal dynamics and the implicit component captures long-term behavioral patterns.
Detailed Evaluation	This paper presents a novel dual conditional diffusion model for sequential recommendation. The authors highlight the necessity of leveraging the historical behaviors as explicit guidance for better diffusion process. In the forward process, the whole user history and the target item are jointly noised together. Then, a self-attention mechanism is utlized to extract high-level implicit user preference from this sequential noisy embeddings. Afterwards, the history item embeddings are explicitly exploited through a cross-attention mechanism for further denoising enhancement. The extensive experiments over three real-world datasets well demonstrate the superiority of the proposed solution.

This paper is well written and easy to follow. The model designs are intuitive and reasonable. As mentioned above, the story of explicit and implicit conditioning is far beyond perfect. The authors should polish the corresponding part carefully and precisely. Overall, this paper would be accepted.
Overall rating
3: Weak Accept (Probable accept, unless convinced otherwise)
Reviewer's confidence
4: Expert: I have published papers on this topic.
Metareview
Metareview
Title	Dual Conditional Diffusion Models for Sequential Recommendation
Authors	Hongtao Huang, Chengkai Huang, Tong Yu, Xiaojun Chang, Wen Hu, Julian McAuley and Lina Yao
Text	This is the meta-review that summarizes the reviews and discussions. This paper introduces a diffusion-based framework for sequential recommendation. To balance contextual representation learning with fine-grained signal guidance, it jointly leverages implicit and explicit conditioning throughout the diffusion process. All reviewers agree that the idea of unifying implicit and explicit conditioning in diffusion-based sequential recommendation is interesting and novel. The model designs are intuitive and reasonable. The experiments are extensive and convincing. The reviewers also pointed out that formal definitions for the implicit and explicit conditioning are missing. The authors should revise the paper according to the review comments in the final version.