Abstract:Recent advancements in diffusion models have shown promising results in sequential recommendation (SR). However, current diffusion-based methods still exhibit two key limitations. First, they implicitly model the diffusion process for target item embeddings rather than the discrete target item itself, leading to inconsistency in the recommendation process. Second, existing methods rely on either implicit or explicit conditional diffusion models, limiting their ability to fully capture the context of user behavior and leading to less robust target item embeddings. In this paper, we propose the Dual Conditional Diffusion Models for Sequential Recommendation (DCRec), introducing a discrete-to-continuous sequential recommendation diffusion framework. Our framework introduces a complete Markov chain to model the transition from the reversed target item representation to the discrete item index, bridging the discrete and continuous item spaces for diffusion models and ensuring consistency with the diffusion framework. Building on this framework, we present the Dual Conditional Diffusion Transformer (DCDT) that incorporates the implicit conditional and the explicit conditional for diffusion-based SR. Extensive experiments on public benchmark datasets demonstrate that DCRec outperforms state-of-the-art methods.

What problem does this paper attempt to address?

The problems that this paper attempts to solve mainly focus on two aspects: 1. **Inconsistency between discrete target items and continuous diffusion processes**: - Current diffusion - model - based methods in recommendation systems usually implicitly model the embeddings of target items rather than directly modeling the discrete target items themselves. This leads to consistency problems in the recommendation process. Specifically, in the existing methods, during the reverse denoising process, the recommended items are determined by calculating the similarity between the reverse target item representation and the candidate item embeddings. This step is deterministic and discrete, which is inconsistent with the continuous generation flow of the diffusion model. This inconsistency results in a mismatch between the optimization directions of the ranking loss of the recommendation task and the denoising loss of the diffusion model, thus affecting the recommendation performance. 2. **Limitations of conditional diffusion models**: - Existing diffusion - model - based recommendation methods can be divided into two categories: implicit - conditional methods and explicit - conditional methods. Implicit - conditional methods integrate users' historical behaviors as implicit features into the generation of target items in each diffusion step, while explicit - conditional methods use historical behaviors as explicit guidance to directly affect the denoising steps in the diffusion process. However, both methods have their inherent limitations: - Implicit - conditional methods may over - simplify complex user behaviors. By compressing historical behaviors into a single vector, important sequential and temporal dynamic information may be lost. - Explicit - conditional methods, although providing detailed conditional information, are more sensitive to noise and irrelevant behaviors and may affect the recommendation quality. To solve the above problems, the authors propose a dual - conditional diffusion model (DCRec), which introduces a sequence recommendation diffusion framework from discrete to continuous and combines implicit and explicit conditional strategies to improve the accuracy and robustness of the recommendation system. Specifically, DCRec solves these problems in the following ways: - **Diffusion framework from discrete to continuous**: - A complete Markov chain is introduced to model the transformation from the reverse target item representation to the discrete item index, ensuring the consistency of the diffusion model. - A unified optimization objective is defined, taking into account both the ranking task and the diffusion task. - **Dual - conditional diffusion transformer (DCDT)**: - It combines implicit and explicit conditions. The global user interests and long - term behavior patterns are captured through implicit conditions, and the sequential complexity and temporal dynamics are preserved through explicit conditions. - Conditional layer normalization (CondLN), multi - head self - attention mechanism (SA), and cross - attention mechanism (CA) are adopted to handle user interaction histories in a more adaptive and context - aware manner. Through these improvements, the experimental results of DCRec on multiple publicly available benchmark datasets show that its performance is better than the existing state - of - the - art methods.

Dual Conditional Diffusion Models for Sequential Recommendation

DiffuRec: A Diffusion Model for Sequential Recommendation

Sequential Recommendation with Diffusion Models

Discrete Conditional Diffusion for Reranking in Recommendation

Semantic-Aware Diffusion Model for Sequential Recommendation

Generative Diffusion Models for Sequential Recommendations

EDiffuRec: An Enhanced Diffusion Model for Sequential Recommendation

Breaking Determinism: Fuzzy Modeling of Sequential Recommendation Using Discrete State Space Diffusion Model

Diff4Rec: Sequential Recommendation with Curriculum-scheduled Diffusion Augmentation

Diffusion Augmentation for Sequential Recommendation

DimeRec: A Unified Framework for Enhanced Sequential Recommendation via Generative Diffusion Models

SeeDRec: Sememe-based Diffusion for Sequential Recommendation

Bridging User Dynamics: Transforming Sequential Recommendations with Schrödinger Bridge and Diffusion Models

Diffusion Recommendation with Implicit Sequence Influence

Diffusion Recommender Model

Multimodal Conditioned Diffusion Model for Recommendation

Dual Sequential Prediction Models Linking Sequential Recommendation and Information Dissemination

Disentangling Past-Future Modeling in Sequential Recommendation Via Dual Networks

Diffusion-based Contrastive Learning for Sequential Recommendation

A novel diffusion recommendation algorithm based on multi-scale cnn and residual lstm

Conditional Denoising Diffusion for Sequential Recommendation