Feature selection: Key to enhance node classification with graph neural networks

Abstract

Graphs help to define the relationships between entities in the data. These relationships, represented by edges, often provide additional context information which can be utilised to discover patterns in the data. Graph Neural Networks (GNNs) employ the inductive bias of the graph structure to learn and predict on various tasks. The primary operation of graph neural networks is the feature aggregation step performed over neighbours of the node based on the structure of the graph. In addition to its own features, for each hop, the node gets additional combined features from its neighbours. These aggregated features help define the similarity or dissimilarity of the nodes with respect to the labels and are useful for tasks like node classification. However, in real-world data, features of neighbours at different hops may not correlate with the node's features. Thus, any indiscriminate feature aggregation by GNN might cause the addition of noisy features leading to degradation in model's performance. In this work, we show that selective aggregation of node features from various hops leads to better performance than default aggregation on the node classification task. Furthermore, we propose a Dual-Net GNN architecture with a classifier model and a selector model. The classifier model trains over a subset of input node features to predict node labels while the selector model learns to provide optimal input subset to the classifier for the best performance. These two models are trained jointly to learn the best subset of features that give higher accuracy in node label predictions. With extensive experiments, we show that our proposed model outperforms both feature selection methods and state-of-the-art GNN models with remarkable improvements up to 27.8%.

DualNetGO: A Dual Network Model for Protein Function Prediction via Effective Feature Selection

“…The architecture of DualNetGO and the training procedure including learning rates and weight decays are based on previous work (Maurya et al ., 2023) or by manual search (Supplementary Table S4,S5). Overall, as the Classifier and the Selector are trained alternately in each epoch of Stage 1 and 2, the Classifier is trained for E 1 + E 2 + E 3 epochs, and the Selector is trained for E 1 + E 2 epochs.…”

“…Recently, Maurya et.al. proposed a feature selection strategy to handle heterogenous graph data, where features of neighbors at different hops may not correlate with node features, which hampers the performance of classical graph neural network (GNN) models on node classification tasks (Maurya et al ., 2023). Their proposed method intelligently determined a suitable combination of features derived from the same graph.…”

Nonlinear Analysis of Bearing Characteristics of Stiffened Deep Cement Mixing Piles under Vertical Loading

“…Whereas many related research studies have been conducted earlier, most of these are based on equal-core piles, and simplified analysis of short-core composite piles is still rare. Today, deep learning [18][19][20] and artificial intelligence [21,22] provide a unique opportunity for predicting the axial force field of piles. Consequently, structural engineering is predictable due to deep learning's specific ability to handle complex nonlinear structural systems under various conditions.…”

Experimental Study on Small-Scale Shake Table Testing of Cable-Stiffened Single-Layer Spherical Latticed Shell

“…In recent years, the rapid development of neural networks and deep learning algorithms has provided forward-looking insights for the optimization of reticulated shell structures. These algorithms offer a variety of approaches for multi-parameter and multi-angle optimization, serving as valuable references [30][31][32][33][34].…”