Lightweight yet Fine-grained: A Graph Capsule Convolutional Network with Subspace Alignment for Shared-account Sequential Recommendation

TL;DR

This work tackles Shared-account Sequential Recommendation (SSR), where multiple latent users share an account and crafting accurate next-item predictions requires disentangling fine-grained interaction ownership. It introduces LightGC$^2$N, a lightweight framework that combines a Graph Capsule Convolutional Network (GC$^2$N) with Subspace Alignment (SA) to capture per-interaction ownership and cluster latent-user preferences, while using linear attention and dynamic routing to maintain efficiency. Key contributions include a capsule-based graph representation for interactions, an account-level routing mechanism, a low-rank subspace clustering with a contrastive objective, and an end-to-end prediction objective that merges refined sequence and account embeddings. Empirically, LightGC$^2$N outperforms nine state-of-the-art SSR methods and uses fewer parameters with faster training on four real-world datasets, supporting practical deployment on resource-constrained devices and enabling more accurate, scalable SSR in diverse settings.

Abstract

Shared-account Sequential Recommendation (SSR) aims to provide personalized recommendations for accounts shared by multiple users with varying sequential preferences. Previous studies on SSR struggle to capture the fine-grained associations between interactions and different latent users within the shared account's hybrid sequences. Moreover, most existing SSR methods (e.g., RNN-based or GCN-based methods) have quadratic computational complexities, hindering the deployment of SSRs on resource-constrained devices. To this end, we propose a Lightweight Graph Capsule Convolutional Network with subspace alignment for shared-account sequential recommendation, named LightGC$^2$N. Specifically, we devise a lightweight graph capsule convolutional network. It facilitates the fine-grained matching between interactions and latent users by attentively propagating messages on the capsule graphs. Besides, we present an efficient subspace alignment method. This method refines the sequence representations and then aligns them with the finely clustered preferences of latent users. The experimental results on four real-world datasets indicate that LightGC$^2$N outperforms nine state-of-the-art methods in accuracy and efficiency.

Figures (6)

• Figure 1: An example to illustrate the shared-account sequential recommendation scenario. $\{ v_1, v_2, \dots, v_6\}$ are the historical behaviors in the hybrid sequence.
• Figure 2: Framework of LightGC$^2$N, where $A_1$ and $A_2$ represent two shared accounts, and $\{ I_1, I_2, \dots, I_6\}$ denote the historical interactions that compose the hybrid sequences for these accounts.
• Figure 3: Comparison of time consumption and parameter scale between LightGC$^2$N and competitive SSR methods.
• Figure 4: Impact of hyper-parameters $\alpha$, $\beta$ and $\gamma$ on HV-E and HA-M.
• Figure 5: Comparison of time consumption and parameter scale between LightGC$^2$N and its variants.