본문 바로가기
  • Home

Improving RecFormer with Sliding Window-Based Data Augmentation

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2026, 31(6), pp.45~53
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : April 21, 2026
  • Accepted : June 3, 2026
  • Published : June 30, 2026

Sungho Choi 1 Minho Kim 2 Namgyu Kim 1

1국민대학교
2주식회사 에크네

Accredited

ABSTRACT

With the rapid growth of the online content market, personalized recommendation systems have become increasingly important, yet Cold-start issues caused by data scarcity remain a critical challenge in new IP content environments. While existing ID-based sequential recommendation models fail to leverage textual attributes, RecFormer—built on the Longformer architecture—also suffers from performance degradation when training data is sparse. This paper proposes a Sliding Window-based data augmentation method for RecFormer, generating multiple training samples from a single user interaction sequence. Experiments on the Amazon Scientific Review dataset show that the proposed method consistently outperforms the baseline RecFormer across all metrics, with greater gains observed under more severe data sparsity conditions.

Citation status

* References for papers published after 2024 are currently being built.