A Survey on Cross-Domain Sequential Recommendation

TL;DR

Cross-domain sequential recommendation (CDSR) addresses data sparsity by modeling user interactions across multiple domains and time. The paper formulates CDSR with a four-dimensional tensor $\Gamma \in \mathbb{R}^{n\times m\times s\times k}$ and provides a dual lens (macro and micro) to categorize fusion structures and learning technologies, including input representations, fusion bridges, and auxiliary learning techniques. It surveys a spectrum of methods from one-level to multi-level fusion structures and from RNN/attention/GNN baselines to transfer and contrastive learning strategies, illustrating how cross-domain signals can be aligned and amplified. The review also catalogs public datasets and representative results, and it outlines critical future directions such as multi-domain simultaneous improvement, heterogeneous information fusion, privacy preservation, fairness, interpretability, and the integration of advanced technologies like LLMs for CDSR.

Abstract

Cross-domain sequential recommendation (CDSR) shifts the modeling of user preferences from flat to stereoscopic by integrating and learning interaction information from multiple domains at different granularities (ranging from inter-sequence to intra-sequence and from single-domain to cross-domain). In this survey, we first define the CDSR problem using a four-dimensional tensor and then analyze its multi-type input representations under multidirectional dimensionality reductions. Following that, we provide a systematic overview from both macro and micro views. From a macro view, we abstract the multi-level fusion structures of various models across domains and discuss their bridges for fusion. From a micro view, focusing on the existing models, we first discuss the basic technologies and then explain the auxiliary learning technologies. Finally, we exhibit the available public datasets and the representative experimental results as well as provide some insights into future directions for research in CDSR.

Figures (6)

• Figure 1: Illustration of CDSR.
• Figure 2: A visualization of dimensionality reduction for a four-dimensional data tensor in CDSR scenarios. We represent the dimensions contained in the data using wireframes and simulate tensors using blocks. The colored blocks indicate records of user interaction with items within the domain at a given moment. In this case, both domains share the same set of users, but there is no overlap in the items they interact with.
• Figure 3: The overview of multi-level fusion structures that are divided into three levels. "A" and "B" represent domain A and domain B, respectively, and "Hybrid" denotes a combination of two domains in chronological order.
• Figure 4: Examples of building cross-domain bridges relying on different information.
• Figure 5: A schematic overview of the key technical framework. The color of the arrows represents the output after passing through the components represented by different colors. While models represented via graph structures require encoding with GNN, the relationship between RNN and attention can be used in parallel or an alternating fashion.