Feature Analysis for Sequential Recommender Systems Using Transformer-Based Architectures
2023 Innovations in Intelligent Systems and Applications Conference, ASYU 2023, Sivas, Türkiye, 11 - 13 Ekim 2023, (Tam Metin Bildiri)
- Yayın Türü: Bildiri / Tam Metin Bildiri
- Doi Numarası: 10.1109/asyu58738.2023.10296691
- Basıldığı Şehir: Sivas
- Basıldığı Ülke: Türkiye
- Anahtar Kelimeler: feature analysis, sequential & session-based recommender systems, transformers
- Boğaziçi Üniversitesi Adresli: Evet
Özet
Recommender systems assist users by suggesting relevant items among a large collection of items. Sequential and session-based recommender systems make recommendations based on the order of items users interact with. These systems take into account varying tastes of the users and can work well for short interaction sequences. In this paper, we make a comprehensive feature analysis in session-based recommendation. We divide the features into different groups and analyze the effects of both the feature groups and the individual features on the performance of the recommender system. We employ the Transformers4Rec framework and three datasets. The results show that time-based features contain the most salient information and using short sequences of past data yields better recommendations.