Following the application of Deep Learning to graphic data, Graph Neural Networks (GNNs) have become the dominant method for Node Classification on graphs in recent years. To assign nodes with preset labels, most GNNs inherit the end-to-end way of Deep Learning in which node features are input to models while labels of pre-classified nodes are used for supervised learning. However, while these methods can make full use of node features and their associations, they treat labels separately and ignore the structural information of those labels. To utilize information on label structures, this paper proposes a method called 3ference that infers from references with differences. Specifically, 3ference predicts what label a node has according to the features of that node in concatenation with both features and labels of its relevant nodes. With the additional information on labels of relevant nodes, 3ference captures the transition pattern of labels between nodes, as subsequent analysis and visualization revealed. Experiments on a synthetic graph and seven real-world graphs proved that this knowledge about label associations helps 3ference to predict accurately with fewer parameters, fewer pre-classified nodes, and varying label patterns compared with GNNs.
The development of Internet of Things systems (IoTs) and 5G technology has allowed image and text information to be collected and spread at an unprecedentedly high speed. To improve the data processing capabilities of IoTs, the semantic relations between images and text should be extracted efficiently and accurately. Therefore, to reduce the enormous semantic differences between images and text, existing methods introduce consensus knowledge graphs into image–text matching tasks. However, these methods result in noisy edges during the graph construction stage and overlook detailed knowledge extraction, leading to reduced performance in semantic matching. In this article, a two-layer heterogeneous knowledge graph network is proposed to solve the above problems. The proposed model incorporates category knowledge and local knowledge for improved data representation. Specifically, a category-based hierarchical knowledge graph is constructed to learn representations of knowledge concepts through a hierarchical correlation graph embedding (HCGE) module. Then, a globally guided local attention (GLA) module is used to extract fine-grained local knowledge. Finally, the similarity between the input image and text is calculated based on knowledge-fused features to complete the matching process. Extensive experiments show that the proposed model can learn more effective knowledge features to improve the efficacy of image–text matching in.