Enhancing effective connection with link prediction for session-based recommendation

The core objective of session-based recommendation (SBR) is to precisely capture user interest preferences from massive candidate items by modeling short-term behavioral patterns. Current approaches predominantly rely on global features or heuristic rules to restructure session graphs for enhanced d...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Neurocomputing (Amsterdam) Jg. 653; S. 131218
Hauptverfasser: Wang, Hui, You, Junjie, Yi, Jianbing, Liu, Junyan, Huang, Runjie
Format: Journal Article
Sprache:Englisch
Veröffentlicht: Elsevier B.V 07.11.2025
Schlagworte:
Information enhancement
Link prediction
Session-based recommendation
Variational graph autoencoder
Link prediction
Information enhancement
Variational graph autoencoder
Session-based recommendation
ISSN:0925-2312
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:The core objective of session-based recommendation (SBR) is to precisely capture user interest preferences from massive candidate items by modeling short-term behavioral patterns. Current approaches predominantly rely on global features or heuristic rules to restructure session graphs for enhanced data representation, yet face a critical limitation: ineffective modeling of implicit item transition patterns within individual sessions. Compounding this issue, while link prediction techniques have demonstrated proven effectiveness in pattern mining, their systematic integration into session recommendation remains underexplored. To address the above issues, this paper proposes the Enhancing effective connection with link prediction for session-based recommendation (ECLP-SR) framework, which introduces link prediction as an information enhancer for session recommendation. Our method employs a Variational Graph Autoencoder (VGAE) to disentangle higher-order item interaction patterns from historical sessions, generating probabilistic transition distributions through latent space inference. This enables quantitative identification of high-frequency, semantically meaningful connections. Unlike the traditional static graph construction method, we design a dynamic weight adjustment mechanism to reconstruct the edge weight distribution of the session graph based on edge importance, thus enhancing the data-driven semantic characterization capability. Experiments on three benchmark datasets show that ECLP-SR improves the semantic distinctiveness of the session graph structure and enhances the model recommendation performance. The source code is available at: https://github.com/Typejunjie/code2. •We introduce link prediction into SBR to learn session history interactions.•Enhancing information by reconstructing connection weights using the VGAE model.•ECLP-SR surpasses state-of-the-art SBR models in experimental evaluations.
ISSN:0925-2312
DOI:10.1016/j.neucom.2025.131218