Aug 2026· Journal of King Saud University: Computer and Information Sciences· Vol 38· 0 citations· 40 references
TL;DR
SimGAT, a structure-aware graph attention model built on SimRank-derived structural embeddings, is proposed, which computes structural similarity in the SimRank2Vec embedding space and injects it as a topological prior into the graph attention mechanism, enabling neighborhood aggregation to be jointly guided by node attributes and global structural relationships.
Abstract
Graph neural networks (GNNs) have achieved strong performance in graph representation learning by integrating node attributes with local topology. However, most existing GNNs rely on local message passing, limiting their ability to capture global structural similarity and higher-order topological dependencies. Although SimRank offers a classical measure of structural-context similarity, its pairwise scalar outputs and quadratic computational cost make direct integration into node-level GNN frameworks impractical. Moreover, conventional SimRank suffers from expressive limitations: zero similarity in directed graphs and structural discrimination collapse in undirected or highly symmetric structures. To address these challenges, we propose SimGAT, a structure-aware graph attention model built on SimRank-derived structural embeddings. We first introduce SimRank2Vec, which converts multi-order SimRank propagation relationships into compact node-level representations through matrix factorization. The resulting embeddings alleviate the expressive limitations of conventional SimRank while providing a scalable structural representation for GNNs. SimGAT then computes structural similarity in the SimRank2Vec embedding space and injects it as a topological prior into the graph attention mechanism, enabling neighborhood aggregation to be jointly guided by node attributes and global structural relationships. Experiments on nine benchmark datasets, namely TwitchPT, TwitchRU, Texas, Cora, Citeseer, Photo, ACM, DBLP, and Film, demonstrate that SimRank2Vec improves structural representation quality, and that SimGAT achieves competitive and stable performance across node classification, community detection, and node representation visualization. The code is available at https://github.com/Chengda217/SimGAT.
A diffusion-enhanced inductive link prediction framework that combines Graph Diffusion Convolution (GDC), structural node descriptors, and neighborhood aggregation from GraphSAGE is proposed that achieves higher accuracy than the other models on the benchmark datasets.
Semantic similarity between nodes in graphs with textual attributes is crucial for numerous graph analytics tasks, such as clustering, inference, and latent relationship detection. Traditional structure-based similarity measures capture only topological information and fail to exploit the semantic content embedded in node descriptions. In this work, we propose a transformer-based semantic similarity framework that leverages contextual embeddings from pretrained Sentence-Transformer models to represent node textual content in a high-dimensional continuous space. Cosine similarity between node embeddings quantifies semantic relatedness, enabling the identification of semantically coherent node pairs and latent clusters, even when nodes are structurally distant or described with different vocabularies. We validate the approach on synthetic graphs inspired by investigative scenarios from the FALCON project and compare its performance against structure-only baselines such as SimRank and node2vec. Results show that the transformer-based method effectively captures hidden semantic relationships, outperforming structural approaches in precision, MAP, and cluster purity, demonstrating the potential of deep contextual embeddings to enrich graph analysis with latent semantic understanding.
Marco San Biagio, Marialuna De Tommaso, Ernesto La Mattina· 2026 6th International Confe...· 0 citations
Graph classification plays a central role in many scientific disciplines. While classical kernel-based methods and graph neural networks achieve strong predictive performance, they often require substantial computational resources. Hyperdimensional Computing (HDC) has recently emerged as an efficient and noise-resilient alternative, providing lightweight models that are attractive for resource-constrained settings. Within this context, GraphHD is a representative HDC-based approach for graph classification; however, its encoding process can become costly on large graphs and its standard configuration relies on a single centrality choice (PageRank) for node-to-hypervector assignment.
In this work, we go beyond PageRank in GraphHD by systematically evaluating alternative centrality measures (degree, closeness, betweenness, Katz, and eigenvector) and by introducing two new encoding variants. GraphHD-Level preserves quantitative structural information by mapping centrality values to level-hypervectors, whereas GraphHD-Order simplifies the algorithm by eliminating edge encoding and aggregating node hypervectors directly. Experiments on six widely used benchmarks from cheminformatics and bioinformatics (MUTAG, ENZYMES, PROTEINS, DD, NCI1, and PTC\_FM) show that replacing PageRank with alternative centralities yields similar F1-scores while offering notable runtime savings, and that GraphHD-Order remains competitive with the original GraphHD baseline while providing consistent speedups in encoding time.
While signed social recommendation has shown great potential by modeling both trust and distrust relations, its effectiveness is often hindered by structural noise and data sparsity. In this work, we first identify a fundamental inconsistency across the structural, propagation, and semantic layers of existing models, which leads to biased representations learned from sparse or noisy datasets. Furthermore, we observe that most existing methods treat the observed graph as fixed, failing to bridge the gap between noisy topologies and reliable social semantics. To address these issues, we propose a unified framework named SSC-Loop that treats signed social recommendation as the maximization of structural consistency. SSC-Loop includes three dedicated modules: ESA-DA for structural consistency, a P/N/O propagation mechanism for propagation consistency, and a contrastive learning objective for semantic consistency. Experiments on Epinions demonstrate that SSC-Loop achieves strong performance on explicit signed social rating prediction, while auxiliary results on Slashdot under a derived link-existence setting further suggest its ability to exploit signed social structures. Source code is available at https://github.com/Refrainwww/SSC-Loop.