상세 보기
A Comparative Study of Sequence Length in Sequential Recommender Systems
- Kang, Hyonjun;
- Yu, Hyunwook;
- Choe, Taeyoung;
- Kim, Mucheol
SCOPUS
0초록
Sequential recommendation models have become increasingly prominent due to their ability to capture the temporal dynamics of user behavior. While prior work has focused on architectural innovations and benchmark performance, limited attention has been given to how these models perform under varying sequence lengths and data sparsity conditions. This paper systematically evaluates two representative transformer-based models SASRec and BERT4Rec. Our results reveal that both models benefit significantly from longer sequences in sparse settings, with BERT4Rec achieving higher peak performance but exhibiting greater sensitivity to input length. In dense environments, even short sequences yield strong performance due to rich global patterns. This study provides actionable insights for deploying sequential recommenders in real-world systems.
키워드
- 제목
- A Comparative Study of Sequence Length in Sequential Recommender Systems
- 저자
- Kang, Hyonjun; Yu, Hyunwook; Choe, Taeyoung; Kim, Mucheol
- 발행일
- 2025
- 유형
- Conference Paper
- 저널명
- International Conference on ICT Convergence
- 페이지
- 892 ~ 894
- 언어
- ENG
- 출판사
- IEEE Computer Society
- 발행국가
- 미국
- 분량
- 3 페이지
- ISSN
- E 2162-1241
P 2162-1233