Revisiting the Assumptions of Shared-Account Sequential Recommendation

초록

Shared-account sequential recommendation (SSR) has been proposed to address the limitation of sequential recommendation (SR) models under shared-account environments. In this paper, we conduct a dataset-centric study to examine how shared-account environments affect SR models and which characteristics of shared-account data drive recommendation performance. By comparing representative SR models on matched single-user and shared-account datasets, we show that performance degradation depends on model architecture and is mitigated by bidirectional contextual modeling. We further establish an analysis framework covering five key shared-account characteristics - group size, sequence length, user similarity, sparsity, and item popularity - and analyze their influence on recommendation performance. Our findings provide support for the motivation of SSR. The code and experimental result tables used in this study are available at the following GitHub link: https://github.com/yejin23/Recsys2026 © 2026 Copyright held by the owner/author(s).

키워드