Towards Robust Cross-Domain Recommendation with Joint Identifiability of User Preference

TL;DR

A hierarchical user preference modeling framework that organizes user representations by the neural network encoder's depth, allowing separate treatment of shallow and deeper subspaces is introduced, highlighting the importance of joint identifiability in achieving robust CDR.

Abstract

Recent cross-domain recommendation (CDR) studies assume that disentangled domain-shared and domain-specific user representations can mitigate domain gaps and facilitate effective knowledge transfer. However, achieving perfect disentanglement is challenging in practice, because user behaviors in CDR are highly complex, and the true underlying user preferences cannot be fully captured through observed user-item interactions alone. Given this impracticability, we instead propose to model {\it joint identifiability} that establishes unique correspondence of user representations across domains, ensuring consistent preference modeling even when user behaviors exhibit shifts in different domains. To achieve this, we introduce a hierarchical user preference modeling framework that organizes user representations by the neural network encoder's depth, allowing separate treatment of shallow and deeper subspaces. In the shallow subspace, our framework models the interest centroids for each user within each domain, probabilistically determining the users' interest belongings and selectively aligning these centroids across domains to ensure fine-grained consistency in domain-irrelevant features. For deeper subspace representations, we enforce joint identifiability by decomposing it into a shared cross-domain stable component and domain-variant components, linked by a bijective transformation for unique correspondence. Empirical studies on real-world CDR tasks with varying domain correlations demonstrate that our method consistently surpasses state-of-the-art, even with weakly correlated tasks, highlighting the importance of joint identifiability in achieving robust CDR.

Figures (6)

• Figure 1: Bridge method (left) and Alignment method (right).
• Figure 2: Overview of CIDER. It separates the user representations into shallow and deep subspaces corresponding to the depth of the neural encoder layers. The shallow subspace encodes domain-irrelevant features, enforced to be aligned between domains, while the deep subspace models domain-relevant factors, where a shared portion and variant portions are further identified by a causal data-generating structure.
• Figure 3: Concept of CPA. The discrepancy between domains is determined by the distance between their respective centroids. We embed shallow user representations into a latent space $P(\cdot)$ characterized by multivariate Gaussian distributions. Compressing $P(\cdot)$ corresponds to finding the centroids ($\boldsymbol{C}_k^{\boldsymbol{X}}(\cdot), \boldsymbol{C}_k^{\boldsymbol{Y}}(\cdot)$) using a statistical distance ${\mathbf{D}}_p$. The discrepancy between the source and target domains is measured by the Kullback-Leibler divergence $D_{\mathrm{KL}}(\cdot)$ between the centroids. For more complex datasets, multiple centroids are required to capture the full range of variability within the data (\ref{['fig:k=3']}).
• Figure 4: The data generation graph (left) and framework (right) for deep subspace disentanglement. The grey circles represent observed variables, while the dashed lines indicate correlations between variables. $R$ denotes the reparameterization trick.
• Figure 5: Impact of different overlapping ratios.

Theorems & Definitions (5)

• Remark 1
• Definition 1: Joint Identifiability
• Proposition 1
• proof
• Remark 2