Search papers, labs, and topics across Lattice.
This paper introduces the Personalized Hypergraph-enhanced Kolmogorov-Arnold Network Transformer (PHKT), which leverages a personalized dynamic hypergraph module to weight item similarities based on users' historical behaviors, addressing the limitations of existing graph methods in heterogeneous semantics and sequence dependency modeling. By integrating a Transformer for temporal modeling and replacing the standard MLP with KAN for enhanced nonlinear response capabilities, PHKT effectively captures user-specific high-order relationships. Experimental results across three real-world datasets demonstrate that PHKT significantly outperforms nine baseline models in predicting target behaviors, underscoring its potential for improving multi-behavior recommendation systems.
PHKT outperforms traditional models by effectively capturing user-specific behaviors and preferences through a novel hypergraph approach and advanced temporal modeling.
In multi-behavior recommendation, auxiliary behaviors such as clicks, add-to-cart, and purchases can provide richer supervisory information for predicting target behaviors. Although existing graph and hypergraph methods are capable of modeling high-order relationships among users, items, and behaviors, they still have limitations in heterogeneous semantics, user-specific weighting, and sequence dependency modeling. While standard Transformers excel at sequence modeling, their shared feedforward mapping struggles to accommodate the differentiated requirements of heterogeneous latent patterns in multi-behavior scenarios. To address this, this paper proposes the Personalized Hypergraph-enhanced Kolmogorov-Arnold Network Transformer (PHKT). Specifically, we design a personalized dynamic hypergraph module that performs behavior-aware weighting of item similarities based on users' historical behavior sequences to capture user-specific heterogeneous high-order relationships. Meanwhile, a Transformer is used as the temporal backbone to model the evolution of short- and long-term preferences, and KAN is introduced to replace the traditional MLP in the feedforward network to enhance fine-grained modeling capability for nonlinear responses to different latent patterns. Experiments on three real datasets, Tmall, RetailRocket, and IJCAI, show that PHKT consistently outperforms nine strong baseline models across multiple evaluation metrics, demonstrating its effectiveness in multi-behavior preference modeling and target behavior prediction.
