A method for link prediction for complex networks

Abstract

The invention provides a link prediction method for complex networks, an end-to-end link prediction model based on graph attention network (GAT), and a batch training method for the model. The key to this model is to learn the attention distribution of network nodes to their surrounding neighbors. The steps of model training and model prediction include: step 1, input the topology structure of the unweighted and undirected homogeneous network; step 2, perform first-order and second-order neighbor sampling on all nodes according to the topology structure of the training set, so that the network Batch; step 3, input the batched training set into the above model to train model parameters; step 4, input the point pair you want to predict, and the model outputs the probability that there is an edge between the point pair. The model described in the present invention has the characteristics of end-to-end. The batch training method makes the model suitable for large-scale complex networks.

Description

Technical field [0001] The present invention relates to deep learning and network science crossains, and specific relates to an end-to-end complex network link prediction model and a batch training method thereof. This model uses the attention mechanism to combine the network topology, which can characterize the network. The method of batch training allows the network to handle link prediction issues of large-scale networks. [0002] technical background [0003] Large-scale complex networks are generally in the real world, such as the World Wide Web, Aviation Network, Online Network and Protein Networks. It is understood that these complex networks are increasingly urgent needs in humans. The study of complex networks belongs to the cross-areas, namely the theoretical research from mathematics and physical perspectives, and also studies with algorithm in combination with computer technology, is one of the research hotspots in the current scientific field. Under normal circumstanc...

AI Technical Summary

Problems solved by technology

Traditional link prediction methods generally regard each part of the network as homogeneous, and do not distinguish the influence of each part on the target node, which is not in line with the actual situation, so there is a certain bottleneck in its prediction effect

Images