Graph anomaly detection identifies abnormal patterns in networks but faces label scarcity and extreme class imbalance. While graph contrastive learning offers unsupervised solutions, existing methods suffer from two limitations: random augmentations break semantic consistency in positive pairs, while naive negative sampling produces trivial contrasts. We propose AC2L-GAD, an Active Counterfactual Contrastive Learning framework addressing both limitations through principled counterfactual reasoning. By combining information-theoretic active selection with counterfactual generation, our approach identifies structurally complex nodes and generates anomaly-preserving positive augmentations alongside hard negative contrasts, while restricting expensive counterfactual generation to a strategically selected subset. This design reduces computational overhead by approximately 65% compared to full-graph counterfactual generation while maintaining detection quality. Experiments on nine benchmark datasets, including real-world financial transaction graphs from GADBench, show that AC2L-GAD achieves competitive or superior performance compared to state-of-the-art baselines, with notable gains in datasets where anomalies exhibit complex attribute-structure interactions.

Identifying influential nodes in complex networks is a fundamental problem with important applications in viral marketing, epidemic control, and social influence analysis. A key challenge in this domain is the scarcity of ground-truth labels: computing accurate influence scores typically requires repeated simulations using diffusion models, which is computationally expensive on large-scale graphs. Additionally, existing deep learning models, particularly those based on graph neural networks (GNNs), often incur high inference costs, limiting their practical deployment. To address these challenges, we propose DistillRWGCN, a knowledge distillation framework based on multi-structural views that captures both structural and local topological features for influence prediction. The framework integrates global and local views of the graph using breadth-first search and depth-first search strategies, fuses the resulting embeddings through an attention mechanism, and distills this knowledge into a lightweight TinyGCN student model via contrastive alignment. DistillRWGCN is trained using only a small subset of influence scores and employs a novel hybrid loss that combines pairwise ranking and Spearman correlation to enhance both prediction accuracy and ranking consistency. Extensive experiments on nine real-world datasets demonstrate that DistillRWGCN achieves up to 23.7 % lower error and 12.5 % higher ranking correlation than state-of-the-art baselines, while significantly reducing inference time. These results highlight the mode’s effectiveness in label-scarce and resource-constrained scenarios.

Recommender systems play a critical role in digital platforms by providing personalized suggestions based on user preferences and interactions. While Graph Neural Networks (GNNs) have improved recommendation accuracy by modeling complex user-item relationships, existing systems often struggle to integrate heterogeneous data sources, including textual reviews, item attributes, and user profiles, into a unified and explainable framework. To address these limitations, we present ExpLMGCN, an explainable multi-modal recommender system that fuses user-item interactions, review embeddings, item features, and user profiles using GNNs and pre-trained language models. Fine-grained user preferences are captured by extracting opinion-aspect pairs (OAs) from reviews and representing them as User–OA and Item–OA bipartite graphs. A contrastive learning framework aligns multi-modal embeddings into a shared latent space, while explanations are generated by ranking OAs based on preference similarity, sentiment, novelty, and factuality. Multiple candidate explanations are produced via large language models, and the optimal one is selected using semantic alignment, multi-aspect coverage, and a self-consistency feedback loop. Extensive experiments on benchmark datasets demonstrate that ExpLMGCN consistently outperforms state-of-the-art baselines, achieving up to 4.39% improvement in recommendation accuracy across Recall and NDCG metrics. Moreover, ExpLMGCN substantially enhances explanation quality, yielding notable gains across BERTScore, ROUGE, and SBERT, and achieving up to 29.4% improvement in BLEU score. These results highlight the model’s ability to provide not only more accurate recommendations but also more coherent and semantically aligned explanations.

Despite Graph Neural Networks (GNNs) having achieved notable success in link prediction tasks, they continue to struggle to model pairwise relationships effectively. Subgraph-based techniques can be effective in small neighborhoods, but they are not scalable, hindering the incorporation of broader structural dependencies. To address this scalability challenge, this paper proposes Pool Walk Subgraph-Based Learning (PWLP) that combines global and local structural information. Our method leverages a pool-walk strategy, incorporating flexible node hops and prioritizing influential nodes within subgraphs, thus capturing both proximal and distant relationships. This architecture effectively integrates structural and node-based features, while mitigating the computational overhead of converting subgraphs to line-graphs by restricting subgraph size through informed candidate node selection, providing scalability across diverse graph settings. Comprehensive evaluations on standard benchmark datasets show that PWLP outperforms state-of-the-art approaches.

Graph- and hypergraph-based recommender systems struggle to learn reliable representations under severe user–item sparsity, particularly in cold-start scenarios. This limitation is compounded by the inefficient integration of auxiliary signals such as social trust networks and user reviews. Existing knowledge distillation methods partially mitigate these challenges but are typically limited to a single teacher and information source, restricting exploitation of heterogeneous information.We model knowledge distillation as a representation alignment process across multiple sources, in which user-centric and item-centric dependencies are extracted and transferred as complementary supervisory signals. The proposed MKDH framework employs two specialised teachers that fuse graph-based structural information with semantic representations from pre-trained BERT. Knowledge from both teachers is transferred to a lightweight HGNN-based student through a contrastive objective, while joint training preserves alignment between the teacher representations and the student’s evolving embeddings. Experiments on Yelp, Ciao, and Epinions show that MKDH consistently outperforms ten strong baselines, achieving statistically significant gains ( ) of up to 2.79% in HR@10 and 4.48% in NDCG@10, with robust performance under highly sparse data. Because only the lightweight student runs at inference, MKDH preserves serving-time efficiency, establishing multi-teacher distillation as an effective paradigm for heterogeneous information in recommendation.

Predicting the influence power of nodes in complex networks, particularly in large-scale scenarios, is a fundamental and challenging problem in network analysis. However, labeling nodes based on their influence power requires running computationally intensive models, such as the Susceptible-Infected-Recovered (SIR) model, which becomes prohibitively time-consuming in large networks, severely limiting scalability. To address this limitation, this study proposes an innovative approach based on multi-level knowledge distillation aimed at enhancing prediction accuracy while substantially reducing inference time, even when few labeled nodes are available. Our approach employs a multi-level teacher-student architecture, enabling knowledge transfer from rich labeled networks to networks with a few labeled nodes. Furthermore, the student model is designed to be shallow, with few parameters, ensuring a lightweight and optimized architecture that significantly reduces the inference time. The transferred knowledge includes both soft labels and an adversarial alignment mechanism between teacher and student models. Experimental results obtained over a range of real-world datasets demonstrate significant improvements in predictive accuracy and computational efficiency.

Abstract Identifying meaningful communities in complex networks is essential for uncovering functional structures in social, biological, and technological systems. Despite the success of Nonnegative Matrix Factorization (NMF) methods, most existing approaches fail to simultaneously preserve local structural patterns, enforce clear community separation, and effectively handle the sparsity of real-world graphs. To address these challenges, we propose SPD-SNMTF , a Structural Proximity and Diagonalized Symmetric Nonnegative Matrix Tri-Factorization framework that integrates three key components: (1) a hybrid similarity matrix combining explicit links and latent neighborhood affinities; (2) diagonal dominance constraints to enhance inter-community separation; and (3) graph Laplacian regularization to preserve topological smoothness. This unified approach jointly reconstructs the adjacency matrix while enforcing geometric and sparsity-aware constraints, resulting in more interpretable community detection with stable performance across a wide range of parameter settings. Experiments on eleven real-world datasets demonstrate that SPD-SNMTF consistently outperforms twelve state-of-the-art methods in terms of NMI, ACC, and ARI. Implementation details are available at https://github.com/sohrabi94/SPD-SNMTF .

In this paper, we propose a novel framework to significantly enhance the inference speed and memory efficiency of Hypergraph Neural Networks (HGNNs) while preserving their high accuracy. Our approach utilizes an advanced teacher-student knowledge distillation strategy. The teacher model, consisting of an HGNN and a Multi-Layer Perceptron (MLP), not only produces soft labels but also transfers structural and high-order information to a lightweight Graph Convolutional Network (GCN) known as TinyGCN. This dual transfer mechanism enables the student model to effectively capture complex dependencies while benefiting from the faster inference and lower computational cost of the lightweight GCN. The student model is trained using both labeled data and soft labels provided by the teacher, with contrastive learning further ensuring that the student retains high-order relationships. This makes the proposed method efficient and suitable for real-time applications, achieving performance comparable to traditional HGNNs but with significantly reduced resource requirements.

This paper proposes SHARP-Distill (\textbf{S}peedy \textbf{H}ypergraph \textbf{A}nd \textbf{R}eview-based \textbf{P}ersonalised \textbf{Distill}ation), a novel knowledge distillation approach based on the teacher-student framework that combines Hypergraph Neural Networks (HGNNs) with language models to enhance recommendation quality while significantly improving inference time. The teacher model leverages HGNNs to generate user and item embeddings from interaction data, capturing high-order and group relationships, and employing a pre-trained language model to extract rich semantic features from textual reviews. We utilize a contrastive learning mechanism to ensure structural consistency between various representations. The student includes a shallow and lightweight GCN called CompactGCN designed to inherit high-order relationships while reducing computational complexity. Extensive experiments on real-world datasets demonstrate that SHARP-Distill achieves up to 68× faster inference time compared to HGNN and 40× faster than LightGCN while maintaining competitive recommendation accuracy.

Multi-label learning tasks involve instances that may belong to multiple categories simultaneously, making feature selection particularly challenging in high-dimensional feature spaces. Existing multi-label feature selection methods often suffer from limitations such as high computational complexity, inadequate handling of feature redundancy, and insufficient modelling of label dependencies. To overcome these challenges, we propose a novel framework called Maximum Relevant Minimum Redundant Multi-Label Feature Selection (MR2MLFS), which integrates a two-layer graph representation with a modified Ant Colony Optimization (ACO) strategy. The first graph layer clusters correlated features using Louvain community detection, while the second constructs a meta-graph to model inter-cluster relationships. ACO then explores this structure, favouring the selection of highly relevant and non-redundant features. To reduce computational overhead, we introduce an information-theoretic metric that estimates both feature-label relevance and feature-feature redundancy, eliminating the need for repeated classifier training during the search. We evaluated the proposed method on ten benchmark multi-label datasets using several multi-label classifiers. Experimental results show that the proposed method outperforms six state-of-the-art methods across multiple evaluation metrics, achieving an average relative improvement of 5–12 % while reducing feature dimensionality by up to 80 %. These results confirm the method's robustness, efficiency, and effectiveness in multi-label feature selection.

Recommendation systems are vital tools for helping users discover content that suits their interests. Collaborative filtering methods are one of the techniques employed for analyzing interactions between users and items, which are typically stored in a sparse matrix. This inherent sparsity poses a challenge because it necessitates accurately and effectively filling in these gaps to provide users with meaningful and personalized recommendations. Our solution addresses sparsity in recommendations by incorporating diverse data sources, including trust statements and an imputation graph. The trust graph captures user relationships and trust levels, working in conjunction with an imputation graph, which is constructed by estimating the missing rates of each user based on the user–item matrix using the average rates of the most similar users. Combined with the user–item rating graph, an attention mechanism fine tunes the influence of these graphs, resulting in more personalized and effective recommendations. Our method consistently outperforms state-of-the-art recommenders in real-world dataset evaluations, underscoring its potential to strengthen recommendation systems and mitigate sparsity challenges. • Triplet path GCN captures nonlinear relationships between users and items. • Sparsity is addressed by adding both imputation and social relation graphs. • Imputation matrix is pre-constructed in preprocessing and used during learning. • Attention mechanism defines graph contributions in the embedded representation space. • Extensive experiments validate the method on two datasets in various settings.

This paper introduces an explainable Graph Neural Network (GNN)-based recommender system that integrates user-item interactions and user reviews to enhance recommendation accuracy and interpretability. The proposed method leverages Temporal Convolutional Networks (TCNs) as a language model to encode user reviews into vector representations, capturing temporal dynamics and contextual information. Additionally, it extracts opinion-aspect pairs from reviews, enabling the system to understand specific product features and user sentiments. Bipartite graphs are constructed to represent interactions between users/items and opinion aspects, facilitating the integration of user reviews into the GNN framework. A contrastive learning approach is employed to combine these graphs with TCN-generated review embeddings, enhancing the system's ability to capture complex relationships. Finally, a recommendation strategy is proposed which considers relevant opinion-aspects as explanations for recommendations. The experiments conducted on several benchmarks reveal that our method outperforms its competitors.

In this paper, an individualized intelligent multiple-model technique is proposed to design automatic artificial pancreas (AP) systems for the glycemic regulation of type 1 diabetic patients. At first, using the multiple-model concept, the insulin-glucose regulatory system is mathematically identified by constructing some local models. In this step, trade-offs between the number of local models and the complexity of the overall closed-loop system are made by defining and solving a bi-objective optimization problem. Then, optimal AP systems are designed by tuning a bank of proportional-integral-derivative (PID) controllers via the genetic algorithm (GA). A fuzzy gain scheduling strategy is employed to determine the participation percentages of the PID controllers in the control action. Finally, two safety mechanisms, called insulin on board (IOB) constraint and pump shut-off, are installed in the AP systems to enhance their performance. To assess the proposed AP systems, in silico experiments are performed on virtual patients of the UVA/Padova metabolic simulator. The obtained results reveal that the proposed intelligent multiple-model methodology leads to AP systems with limited hyperglycemia and no severe hypoglycemia.

The major challenge of learning from multi-label data has arisen from the overwhelming size of label space which makes this problem NP-hard. This problem can be alleviated by gradually involving easy to hard tags into the learning process. Besides, the utilization of a diversity maintenance approach avoids overfitting on a subset of easy labels. In this paper, we propose a self-paced multi-label learning with diversity (SPMLD) which aims to cover diverse labels with respect to its learning pace. In addition, the proposed framework is applied to an efficient correlation-based multi-label method. The non-convex objective function is optimized by an extension of the block coordinate descent algorithm. Empirical evaluations on real-world datasets with different dimensions of features and labels imply the effectiveness of the proposed predictive model.

There are several potent measures for mining the relationships among actors in social network analysis. Betweenness centrality measure is extensively utilized in network analysis. However, it is quite time-consuming to compute exactly the betweenness

Mechanisms on automatic discovery of macro actions or skills in reinforcement learning methods are mainly focused on subgoal discovery methods. Among the proposed algorithms, those based on graph centrality measures demonstrate a high performance gai

Reinforcement Learning (RL) is an approach for training agent's behavior through trial-and-error interactions with a dynamic environment. An important problem of RL is that in large domains an enormous number of decisions are to be made. Hence, instead of learning using individual primitive actions, an agent could learn much faster if it could form high level behaviors known as skills. Graph-based approach, that maps the RL problem to a graph, is one of the several approaches proposed to identify the skills to learn automatically. In this paper we propose a new centrality measure for identifying bottleneck nodes crucial to develop useful skills. We will show through simulations for two benchmark tasks, namely, “two-room grid” and “taxi driver” that a procedure based on the proposed measure performs better than the procedure based on closeness and node betweenness centrality.