Time-aware multi-behavior graph network model for complex group behavior prediction

In the multifaceted landscape of social networks, user behaviors manifest in various patterns, contributing to the diversity of group behaviors. Current research on group behavior modeling often limits its focus to single behavioral types, overlooking the interplay among different behaviors. To brid...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Information processing & management 2024-05, Vol.61 (3), p.103666, Article 103666
Hauptverfasser: Yu, Xiao, Li, Weimin, Zhang, Cai, Wang, Jingchao, Zhao, Yan, Liu, Fangfang, Pan, Quanke, Liu, Huazhong, Ding, Jihong, Chen, Dehua
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:In the multifaceted landscape of social networks, user behaviors manifest in various patterns, contributing to the diversity of group behaviors. Current research on group behavior modeling often limits its focus to single behavioral types, overlooking the interplay among different behaviors. To bridge this gap, we introduce Time-aware Multi-behavior Graph Network (TMGN) model. This model integrates heterogeneous graph representation learning to discern patterns in user-item interactions across multiple behaviors, capitalizing on dynamic user preferences through time encoding strategy. Additionally, TMGN harnesses a self-attention multi-behavior fusion network to effectively amalgamate characteristics of diverse behaviors, which can tackle the complex hierarchical dependences among distinct group behaviors. Empirical validation on Yelp and the MovieLens 10M datasets demonstrates that TMGN outperforms the leading baseline model, KHGT, by 5.9 %, 23.93 %, and 8.57 % in HR@5, NDCG@5, and Recall@5 metrics, respectively. The findings offer substantial theoretical and practical insights for predicting group behavior on online platforms.
ISSN:0306-4573
1873-5371
DOI:10.1016/j.ipm.2024.103666