Fig 1.
Overall structure diagram of the MBH-GNN model, illustrating the key components: Graph Construction (A), Neighbor-aware GNN (B), High-hop Interaction Learning (C), and Recommendation Prediction (D).
The diagram provides a high-level overview of the model architecture, aiding understanding before the detailed explanation.
Fig 2.
Structure of the multi-behavior interaction graph.
Fig 3.
Structure of the neighbor-aware heterogeneous graph neural network.
Fig 4.
Structure of the high-hop interaction learning module.
Table 1.
Performance comparison on BeiBei and Tmall datasets.
Table 2.
Comparison of model performance on the BeiBei and Tmall datasets, evaluated using multiple recommendation metrics: NDCG and HR at different K values.
Fig 5.
The comparison of Hit Rate (HR@10) and Normalized Discounted Cumulative Gain (NDCG@10) across varying data sparsity levels on BeiBei and Tmall datasets.
Fig 6.
Performance of various models in cold start scenarios on BeiBei and Tmall datasets.
The horizontal axis represents the number of cold start users/items, while the vertical axis shows the respective performance metrics.
Fig 7.
Visualization of user-item multi-behavior interactions in MBH-GNN.
Heatmaps represent the correlation strength between different behaviors, and bar charts show the relative weights of each behavior. Higher correlations and weights indicate greater influence on recommendation decisions.
Table 3.
Ablation study results on BeiBei and Tmall datasets.
Table 4.
Neighborhood contribution analysis on BeiBei and Tmall datasets. This table illustrates the impact of different neighborhood hops (1-hop, 1+2-hop, 1+2+3-hop) on HR@10 and NDCG@10 for both datasets.