QGCNLP: Hybrid Quantum–classical Graph Convolutional Network based Link Prediction

Recently, graph convolutional networks (GCN) have become very popular due to the fact that they are an effective way to perform convolution on a non-Euclidean space-like graph and that convolution operations on neural networks have produced remarkable results. However, GCN experiences speed and memory limitations on traditional computing platforms as the network’s size and complexity increase. Quantum computing, on the other hand, has proven to be very effective in increasing expressiveness by exploring the high-dimensional Hilbert space for complex correlation, alongside providing extraordinarily high computational parallelism. As a result, there is considerable potential for integrating these two advanced technologies. Therefore, we suggested a novel hybrid Quantum–classical Graph Convolutional network-based Link Prediction (QGCNLP) model for the complex task of Link Prediction. This model applies quantum enhancements in a very unique way; instead of converting all input features to quantum data, it applies quantum circuits to the aggregated features obtained after graph convolution in every GCN layer. The quantum-enhanced aggregated values are passed repeatedly till the readout layer, just like the traditional GCN message passing, and then optimized for minimizing the link prediction loss. This approach boosts the model’s performance without demanding substantial quantum resources, making it ideal for the present NISQ era. We firmly demonstrate that our suggested methods are superior after doing rigorous testing on several static datasets as well as dynamic ones with a number of performance evaluation metrics over varied learning rates, qubit counts, circuit variations, a number of benchmark classical models, and state-of-the-art methods.