Abstract:Graph convolutional networks (GCNs) have attracted increasing attention in various fields due to their significant capacity to process graph-structured data. Typically, the GCN model and its variants heavily rely on the transmission of node features across the graph structure, which implicitly assumes that the graph structure and node features are consistent, i.e., they carry related information. However, in many real-world networks, node features may unexpectedly mismatch with the structural information. Existing GCNs fail to generalize to inconsistent scenarios and are even outperformed by models that ignore the graph structure or node features. To address this problem, we investigate how to extract representations from both the graph structure and node features. Consequently, we propose the multi-channel graph convolutional network (MCGCN) for graphs with inconsistent structures and features. Specifically, the MCGCN encodes the graph structure and node features using two specific convolution channels to extract two separate specific representations. Additionally, two joint convolution channels are constructed to extract the common information shared by the graph structure and node features. Finally, an attention mechanism is utilized to adaptively learn the importance weights of these channels under the guidance of the node classification task. In this way, our model can handle both consistent and inconsistent scenarios. Extensive experiments on both synthetic and real-world datasets for node classification and recommendation tasks show that our methods, MCGCN-A and MCGCN-I, achieve the best performance on seven out of eight datasets and the second-best performance on the remaining dataset. For simpler graph structures or tasks where the overhead of multiple convolution channels is not justified, traditional single-channel GCN models might be more efficient.
engineering, electrical & electronic,computer science, information systems,physics, applied
What problem does this paper attempt to address?
The problem that this paper attempts to solve is that in graph - structured data, when node features are inconsistent with the graph structure, existing graph convolutional networks (GCNs) are unable to effectively extract useful representations. Specifically, in many real - world network data, node features may not match structural information, resulting in poor performance of existing GCNs when processing these inconsistent data, and may even be inferior to models that ignore graph structures or node features. To address this challenge, the paper proposes a multi - channel graph convolutional network (MCGCN), which aims to extract representations from the graph structure and node features respectively, and extract the information shared by both through a joint convolutional channel. Finally, an attention mechanism is used to adaptively learn the importance weights of each channel, thereby improving the performance of the model when processing inconsistent data.
### Main Contributions
1. **Research on Inconsistency**: The paper first studies the inconsistency between the structure and node features, and shows the limitations of existing GCNs in fusing inconsistent structures and features through two examples.
2. **Proposing MCGCN**: The paper proposes a multi - channel graph convolutional network (MCGCN). This method can extract representations from the structure and feature spaces respectively, and extract the information shared by both through a joint convolutional channel. In addition, an attention mechanism is introduced to adaptively fuse these representations.
3. **Experimental Verification**: Extensive experiments on synthetic and real - world datasets show that the proposed MCGCN outperforms existing state - of - the - art methods when processing graph data with inconsistent structures and features, and also shows competitiveness when processing graph data with consistent structures and features.
### Method Overview
1. **Specific Convolutional Channels**: MCGCN uses two specific convolutional channels to extract representations from the feature space and the structure space respectively. Specifically, by generating a feature structure (such as a k - nearest neighbor graph) and the original node features, a two - layer graph convolutional network is used to extract the feature representation; at the same time, through the original network structure and the initial topological features, a two - layer graph convolutional network is used to extract the structure representation.
2. **Joint Convolutional Channels**: In order to capture the common information between the structure and the features, MCGCN designs two joint convolutional channels, which respectively propagate the original features on the structure and propagate the topological features on the feature structure.
3. **Attention Mechanism**: Finally, through the attention mechanism, the importance weights of each channel are adaptively learned, so as to extract the most useful information in downstream tasks.
### Formula Examples
- **Feature Similarity Matrix**:
\[
S_{ij}=\frac{x_i\cdot x_j}{\|x_i\|\|x_j\|}
\]
where \(x_i\) and \(x_j\) represent the feature vectors of nodes \(v_i\) and \(v_j\) respectively.
- **Feature Space Convolution**:
\[
Z_f^{(l)}=\sigma(\tilde{D}_f^{-1/2}\tilde{A}_f\tilde{D}_f^{-1/2}Z_f^{(l - 1)}W_f^{(l)})
\]
where \(\tilde{A}_f = A_f+I\), \(\tilde{D}_f\) is the degree matrix of \(\tilde{A}_f\), \(W_f^{(l)}\) is the weight matrix of the \(l\)-th layer, and \(Z_f^{(0)}=X\).
- **Structure Space Convolution**:
\[
Z_s^{(l)}=\sigma(\tilde{D}^{-1/2}\tilde{A}\tilde{D}^{-1/2}Z_s^{(l - 1)}W_s^{(l)})
\]
where \(\tilde{A}=A + I\), \(\tilde{D}\) is the degree matrix of \(\tilde{A}\), \(W_s^{(l)}\) is the weight matrix of the \(l\)-th layer, and \(Z_s^{(0)}=X_s\).
Through these methods, MCGCN can effectively handle the situation where the graph structure and node features are inconsistent, and improve the robustness and performance of the model.