Betweenness Approximation for Edge Computing with Hypergraph Neural Network

Edge computing is highly demanded to achieve their full potentials Internet of Things (IoT), since various IoT systems have been generating big data facilitating modern latency-sensitive applications. As a basic problem, network dismantling tries to find an optimal set of nodes of which will maximize the connectivity degradation in a network. However, current approaches mainly focus on simple networks modeling only pairwise interactions between two nodes, while higher order groupwise interactions among arbitrary number of nodes are ubiquitous in real world which can be better modeled as hypernetwork. The structural difference between simple network and hypernetwork restricts the direct application of simple network dismantling methods to hypernetwork. Even though some hypernetwork centrality measures such as betweenness can be used for hypernetwork dismantling, they face the problem of balancing effectiveness and efficiency. Therefore, we propose a betweenness approximation-based hypernetwork dismantling method with hypergraph neural network, namely HND. HND trains a transferable hypergraph neural network-based regression model on plenty of generated small-scale synthetic hypernetwork in a supervised way, and utilizes the well-trained model to approximate nodes' betweenness. Extensive experiments on five real hypernetworks demonstrate the effectiveness and efficiency of HND comparing with various baselines.