Search

This survey article presents a comprehensive and conceptual overview of anomaly detection (AD) using dynamic graphs. We focus on existing graph-based AD techniques and their applications to dynamic networks. The contributions of this survey article include the following: (i) a comparative study of existing surveys on AD; (ii) a Dynamic Graph-based anomaly detection (DGAD) review framework in which approaches for detecting anomalies in dynamic graphs are grouped based on traditional machine learning models, matrix transformations, probabilistic approaches, and deep learning approaches; (iii) a discussion of graphically representing both discrete and dynamic networks; and (iv) a discussion of the advantages of graph-based techniques for capturing the relational structure and complex interactions in dynamic graph data. Finally, this work identifies the potential challenges and future directions for detecting anomalies in dynamic networks. This DGAD survey approach aims to provide a valuable resource for researchers and practitioners by summarizing the strengths and limitations of each approach, highlighting current research trends, and identifying open challenges. In doing so, it can guide future research efforts and promote advancements in AD in dynamic graphs. [ABSTRACT FROM AUTHOR]

Financial statement fraud refers to malicious manipulations of financial data in listed companies’ annual statements. Traditional machine learning approaches focus on individual companies, overlooking the interactive relationships among companies that are crucial for identifying fraud patterns. Moreover, fraud detection is a typical imbalanced binary classification task with normal samples outnumbering fraud ones. In this paper, we propose a multi-relational graph convolutional network, named FraudGCN, for detecting financial statement fraud. A multi-relational graph is constructed to integrate industrial, supply chain, and accounting-sharing relationships, effectively encapsulating the multidimensional and complex interactions among companies. We then develop a multi-relational graph convolutional network to aggregate information within each relationship and employ an attention mechanism to fuse information across multiple relationships. The attention mechanism enables the model to distinguish the importance of different relationships, thereby aggregating more useful information from key relationships. To alleviate the class imbalance problem, we present a diffusion-based under-sampling strategy that strategically selects key nodes globally for model training. We also employ focal loss to assign greater weights to harder-to-classify minority samples. We build a real-world dataset from the annual financial statement of listed companies in China. The experimental results show that FraudGCN achieves an improvement of 3.15% in Macro-recall, 3.36% in Macro-F1, and 3.86% in GMean compared to the second-best method. The dataset and codes are publicly available at: https://github.com/XNetLab/MRG-for-Finance.

A regression model based on graph neural network, tailored for electric circuit dynamics prediction is introduced, providing converter performance predictions on converter circuit level and internal parameter variations. Regardless of the number of components or connections present in a converter circuit, the proposed model can be readily scaled to incorporate different converter circuit topologies. Moreover, the model can be used to analyse converter circuits with any number of circuit components and any control parameters variation. To enable the use of machine learning methods and applications, all physical and switching circuit properties such as converter circuits operating in continuous conduction mode or discontinuous conduction mode are accurately mapped to graph representation. Three of the most common converters (Buck, Boost, and Buck-boost) are used as example circuits applied to model and the target is to predict the gain and current ripples in inductor. The model achieves 99.51% on the R 2 measure and a mean square error of 0.0263. [ABSTRACT FROM AUTHOR]

Social network models often struggle to capture and adapt to these changing dynamics, leading to limitations in accuracy, scalability, and customization. This research introduces a novel approach to enhancing social network modelling through the integration of advanced techniques such as neural network-driven customization and scalability. Leveraging Multi-Layer Perceptron (MLP) architectures accommodate the complexity and dynamics inherent in modern social networks. The methodology incorporates Wi-Fi Aware (NAN) technology, facilitating efficient neighbor awareness networking to capture real-time interactions and network topologies. Graph Neural Networks (GNN) is employed to handle the intricate structure of social graphs, optimizing connectivity and information propagation within the network. The power of cloud computing platforms such as Microsoft Azure and serverless frameworks like AWS Lambda is harnessed to ensure scalability and flexibility in the model. Tokenization techniques are utilized for efficient data representation and processing, while Feature Extraction Modules enable the extraction of meaningful features from raw social network data. To address the challenges of decision-making and optimization in dynamic social environments, reinforcement learning algorithms such as Deep Q-Networks (DQN), Proximal Policy Optimization (PPO), and Model-Agnostic Meta-Learning (MAML) are integrated. These algorithms enable adaptive learning and decision-making, enhancing the model's responsiveness to changing network conditions. Through extensive experimentation and evaluation, the efficacy of the approach in achieving significant improvements in social network modelling such as accuracy, scalability, and customization is demonstrated. The overall result indicates that the average accuracy across all techniques demonstrates notable variations, with the highest average accuracy exceeding 84.8% and the lowest falling below 78.4%. [ABSTRACT FROM AUTHOR]

Soil heavy metal contamination has emerged as a global environmental concern, posing significant risks to human health and ecosystem integrity. Hyperspectral technology, with its non-invasive, non-destructive, large-scale, and high spectral resolution capabilities, shows promising applications in monitoring soil heavy metal pollution. Traditional monitoring methods are often time-consuming, labor-intensive, and expensive, limiting their effectiveness for rapid, large-scale assessments. This study introduces a novel deep learning method, SpecMet, for estimating heavy metal concentrations in naturally contaminated agricultural soils using hyperspectral data. The SpecMet model extracts features from hyperspectral data using convolutional neural networks (CNNs) and achieves end-to-end prediction of soil heavy metal concentrations by integrating attention mechanisms and graph neural networks. Results demonstrate that the OR-SpecMet model, which utilizes raw spectral data, achieves optimal prediction performance, significantly surpassing traditional machine learning methods such as multiple linear regression, partial least squares regression, and support vector machine regression in estimating concentrations of lead (Pb), copper (Cu), cadmium (Cd), and mercury (Hg). Moreover, training specialized OR-SpecMet models for individual heavy metals better accommodates their unique spectral-concentration relationships, enhancing overall estimation accuracy while achieving a 20.3 % improvement in predicting low-concentration mercury. The OR-SpecMet method showcases the superior performance and extensive application potential of deep learning techniques in precise soil heavy metal pollution monitoring, offering new insights and reliable technical support for soil pollution prevention and agricultural ecosystem protection. The code and datasets used in this study are publicly available at: https://github.com/zhang2lei/metal.git.

Various studies have been conducted to detect network anomalies. However, because anomaly signals are determined by the pattern characteristics using the dataset, the real-time detection problem continues. Even if there is a signal with an attack sign among the constantly transmitted and received signals, the attack cannot be blocked in advance. Moreover, it appears in many places in a distributed denial-of-service (DDoS) attack, so the real-time defense must be the best option. Therefore, it is necessary first to discover the characteristics and elements regarded as abnormal signals to discover anomalies in real time. Finally, by analyzing the correlation between network data and features, extracting the elements of the anomaly, and analyzing the behavior of the extracted elements in detail, we aim to increase the accuracy of the anomaly. In this study, we used Coburg intrusion detection and KDDCup datasets and analyzed the correlation of elements in the dataset using a graph neural network. The calculated accuracy values of the anomaly detection were 94.5% and 98.85%. [ABSTRACT FROM AUTHOR]

The widespread use of smartphones has brought convenience and connectivity to the fingertips of the masses. As a result, this has paved the way for potential security vulnerabilities concerning sensitive data, particularly by exploiting side-channel attacks. When typing on a smartphone’s keyboard, its vibrations can be misused to discern the entered characters, thus facilitating side-channel attacks. These smartphone hardware sensors can capture such information while users input sensitive data like personal details, names, email addresses, age, bank details and passwords. This study presents a novel Graph Neural Network (GNN) approach to predict side-channel attacks on smartphone keyboards; different GNN architectures were used, including GNN, DeepGraphNet, Gradient Boosting (GB)+DeepGraphNet, Extreme Gradient Boosting (XGB)+DeepGraphNet and K-Nearest Neighbor (KNN)+DeepGraphNet. The proposed approach detects the side channel attack using vibrations produced while typing on the smartphone soft keyboard. The data was collected from three smartphone sensors, an accelerometer, gyroscope, and magnetometer, and evaluated this data using common evaluation measures such as accuracy, precision, recall, F1-score, ROC curves, confusion matrix and accuracy and loss curves. This study demonstrated that GNN architectures can effectively capture complex relationships in data, making them well-suited for analyzing patterns in smartphone sensor data. Likewise, this research aims to fill a crucial gap by enhancing data privacy in the information entered through smartphone keyboards, shielding it from side-channel attacks by providing an accuracy of 98.26%. Subsequently, the primary objective of this study is to assess the effectiveness of GNN architectures in this precise context. Similarly, the GNN model exhibits compelling performance, achieving accuracy, precision, recall, and f1 score metrics that showcase the model’s effectiveness, with the highest values of 0.98, 0.98, 0.98, and 0.98, respectively. Significantly, the metrics mentioned in the study outperform those documented in the previous literature. Overall, the study contributes to the detection of side-channel smartphone attacks, which advances secure data practices.INDEX TERMS Graph neural networks (GNN), keystroke inference, motion sensors, machine learning, smartphone security, side-channel attacks.

This article proposes an integration between a graph framework for circuit representation and a Graph neural network (GNN) model suitable for different machine learning (ML) applications. Furthermore, the paper highlights design steps for tailoring and using the GNN-based ML model for converter performance predictions based on converter circuit level and internal parameter variations. Regardless of the number of components or connections present in a converter circuit, the proposed model can be readily scaled to incorporate different converter circuit topologies and may be used to analyze such circuits regardless of the number of components used or control parameters varied. To enable the use of ML methods and applications, all physical and switching circuit properties including operating mode, components and circuit behavior must be accurately mapped to graph representation. The model scalability to other circuit types and different connections and circuits elements is also tested, while being studied in the most common DC-AC inverter in grid connected systems including filter and filterless configurations. The filtered and filterless DC-AC inverter circuits are used to evaluate the model, scoring $R^{2}$ greater than 99% in most cases and a mean square error (MSE) tending to zero.

Autism spectrum disorder (ASD) poses as a multifaceted neurodevelopmental condition, significantly impacting children's social, behavioral, and communicative capacities. Despite extensive research, the precise etiological origins of ASD remain elusive, with observable connections to brain activity. In this study, we propose a novel framework for ASD detection, extracting the characteristics of functional magnetic resonance imaging (fMRI) data and phenotypic data, respectively. Specifically, we employ recursive feature elimination (RFE) for feature selection of fMRI data and subsequently apply graph neural networks (GNN) to extract informative features from the chosen data. Moreover, we devise a phenotypic feature extractor (PFE) to extract phenotypic features effectively. We then, synergistically fuse the features and validate them on the ABIDE dataset, achieving 78.7% and 80.6% accuracy, respectively, thereby showcasing competitive performance compared to state-of-the-art methods. The proposed framework provides a promising direction for the development of effective diagnostic tools for ASD. [ABSTRACT FROM AUTHOR]

Systemic risk is the risk of infection from an individual financial entity to the financial system due to existing interconnections. Having powerful tools to analyze and predict systemic risk in large financial networks is essential to ensure the stability of the financial system, avoiding the negative externalities derived from the failure of a systemically important financial institution. In this context, Machine Learning (ML) has proved to be a useful tool thanks to its ability to deal with complex relations. However, traditional techniques are limited in their use of the interactions between entities and the network structure, which has been shown to be of great importance for systemic risk. Thus, this work proposes Graph Neural Networks (GNNs) for systemic risk analysis. GNNs use the network structure and feature information to deal with large-scale financial networks, providing the benefits of ML while using all the available information (node inter-relations and node, edge, and graph features). We also present C2R, an approach to reduce the pre-labeling effort for costly systemic risk metrics by pre-labeling into a small number of classes while predicting continuous risk scores. We have tested GNNs against traditional ML in classifying entities by systemic risk importance in two different networks, comparing their generalization capabilities with different amounts of available data. GNNs achieve a 94% and 15% Matthew's Correlation Coefficient (MCC) average percentage increase compared to ML, achieving statistically significant MCC improvements in most scenarios. When combining C2R with GNNs to predict the systemic risk quantile from the class labels, the models achieve statistically significant improvements in the quantile RMSE. From our experiments, we can confirm that GNN models are better suited for systemic risk prediction on financial networks and should be preferred over traditional Machine Learning. The results obtained also confirm the fact that the network structure and the features of the relations (edge features) hold useful information for our task.

This paper proposes a method of transferring physical continuous and switching/converter circuits working in continuous conduction mode (CCM) and discontinuous conduction mode (DCM) to graph representation, independent of the connection or the number of circuit components, so that machine learning (ML) algorithms and applications can be easily applied. Such methodology is generalized and is applicable to circuits with any number of switches, components, sources and loads, and can be useful in applications such as artificial intelligence (AI) based circuit design automation, layout optimization, circuit synthesis and performance monitoring and control. The proposed circuit representation and feature extraction methodology is applied to seven types of continuous circuits, ranging from second to fourth order and it is also applied to three of the most common converters (Buck, Boost, and Buck-boost) operating in CCM or DCM. A classifier ML task can easily differentiate between circuit types as well as their mode of operation, showing classification accuracy of 97.37% in continuous circuits and 100% in switching circuits.

Autism spectrum disorder (ASD) poses as a multifaceted neurodevelopmental condition, significantly impacting children’s social, behavioral, and communicative capacities. Despite extensive research, the precise etiological origins of ASD remain elusive, with observable connections to brain activity. In this study, we propose a novel framework for ASD detection, extracting the characteristics of functional magnetic resonance imaging (fMRI) data and phenotypic data, respectively. Specifically, we employ recursive feature elimination (RFE) for feature selection of fMRI data and subsequently apply graph neural networks (GNN) to extract informative features from the chosen data. Moreover, we devise a phenotypic feature extractor (PFE) to extract phenotypic features effectively. We then, synergistically fuse the features and validate them on the ABIDE dataset, achieving 78.7% and 80.6% accuracy, respectively, thereby showcasing competitive performance compared to state-of-the-art methods. The proposed framework provides a promising direction for the development of effective diagnostic tools for ASD.

Augmented Intelligence of Things empowered by knowledge graph drives cognitive intelligence for smart enterprise management systems (EMS). Knowledge fusion technology can effectively integrate knowledge from different sources, thereby improving the accuracy and richness of the knowledge graph, which is of great significance to the sustainable development of smart EMS. Traditional machine learning methods on graphs face challenges in the fusion of complex and multirelational enterprise knowledge graphs due to inherent defects in relation semantic and local structure information capturing. In order to break through these limitations and improve EMS knowledge graphs, we propose tensor-based graph attention networks for multirelational graph representation learning (MR-GAT), and apply it to the critical tasks in knowledge fusion: Entity and relation alignment. Specifically, we innovatively adopt tensor operations to adequately model the interactions between entities and relations in EMS knowledge graph to learn more accurate representations. Additionally, we propose a relation attention mechanism, which focuses on assigning weights in the process of aggregating local semantic information for relation learning in an EMS knowledge graph. Furthermore, we develop a joint entity and relation alignment framework by utilizing the proposed multirelational graph attention networks to improve the accuracy of knowledge fusion. Experimental evaluations on three datasets present that the proposed approach outperforms the baseline models by about 1.4% on average in terms of the mean reciprocal rank metric, which demonstrates the superior ability of the proposed MR-GAT in representation learning for knowledge fusion in smart EMS. [ABSTRACT FROM AUTHOR]

Graph Neural Networks (GNN) show good performance in relational data classification. However, their contribution to concept learning and the validation of their output from an application domain's and user's perspective have not been thoroughly studied. We argue that combining symbolic learning methods, such as Inductive Logic Programming (ILP), with statistical machine learning methods, especially GNNs, is an essential forward-looking step to perform powerful and validatable relational concept learning. In this contribution, we introduce a benchmark for the conceptual validation of GNN classification outputs. It consists of the symbolic representations of symmetric and non-symmetric figures that are taken from a well-known Kandinsky Pattern data set. We further provide a novel validation framework that can be used to generate comprehensible explanations with ILP on top of the relevance output of GNN explainers and human-expected relevance for concepts learned by GNNs. Our experiments conducted on our benchmark data set demonstrate that it is possible to extract symbolic concepts from the most relevant explanations that are representative of what a GNN has learned. Our findings open up a variety of avenues for future research on validatable explanations for GNNs. [ABSTRACT FROM AUTHOR]

With the rapid growth in access demand for Internet of Things (IoT) devices, effective utilization of spectrum resource has become a key challenge to ensure reliable communications. Traditional dynamic spectrum access methods are inefficient when there are too many device accesses, channel reductions, and channel quality deterioration. In this paper, we propose a dynamic spectrum access method based on a fusion algorithm of graph neural network (GNN) and deep Q network (DQN), improving spectrum access efficiency while maintaining a good access success accuracy. Compared with traditional DQN, the computation time can be reduced by over 35%. Our approach first uses GNN to interact with the environment and predict the state of the IoT spectrum environment. Subsequently, automatic learning and optimization of spectrum access policies are achieved by selecting the mobile IoT user's actions based on these predicted states using the DQN's target network, experience playback, and reinforcement learning techniques. Simulation results show that the system model based on the proposed method can operate with better efficiency than the conventional method while maintaining a good channel access rate and channel quality. [ABSTRACT FROM AUTHOR]

Effective graph analysis methods can reveal the intrinsic characteristics of graph data. However, graph is non-Euclidean data, which leads to high computation and space cost while applying traditional methods. Graph embedding is an efficient method for graph analysis. It converts original graph data into a low-dimensional space and retains key information to improve the performance of downstream tasks such as node classification, link prediction, and node clustering. Different from previous studies, this paper focuses on both static and dynamic graph embedding. Firstly, this paper proposes a universal taxonomy of static and dynamic methods, including matrix factorization based methods, random walk based methods, autoencoder based methods, graph neural networks (GNN) based methods and other embedding methods. Secondly, this paper analyzes the theoretical relevance of static and dynamic methods, and comprehensively summarizes the core strategy, downstream tasks and datasets. Finally, this paper proposes four potential research directions of graph embedding.

The ambient air pollution problem has become more severe as the social economy develops. Abnormal event detection in air quality data can prevent property loss and protect human health. The majority of existing anomaly detection models in air quality data are based on a single variable or a single monitoring station, ignoring spatial correlation and multivariate features of regional air pollutant concentrations. To address the aforementioned issues, this paper proposes a new deep learning approach that combines spatial correlation and temporal correlation of air quality data to detect air quality anomalies. The essential idea of this approach is to use the correlation degree between node information to fuse the spatial correlational feature and temporal correlational feature of air quality data to construct spatiotemporal graph structure data, which are used for anomaly detection in air quality data. In this proposed approach, (1) the weighted adjacency matrix is established to characterize the spatial correlation, and the feature matrix is constructed to characterize the temporal correlation; (2) changes in node information are transformed into changes in edge weights using the correlation degree between node information, and spatiotemporal graph structure data are constructed to characterize the spatiotemporal correlation; and (3) an advanced deep learning model, Context augmented Graph Autoencoder (Con-GAE), is utilized to handle the above spatiotemporal graph structure data and detect abnormal air quality events. The efficiency of the approach is demonstrated by anomaly detection on synthetic test sets produced from real-world datasets.

Graph neural networks (GNN) have been successfully applied in complex tasks in many fields, but recent studies show that GNN is vulnerable to graph adversarial attacks, leading to severe performance degradation. The vulnerability of GNN affects all applications including node classification, link prediction and community detection. Graph adversarial attacks can be implemented efficiently, which brings serious security risks and privacy issues. Graph adversarial defense is dedicated to improving the robustness and generalization of GNN to resist adversarial attacks. Research progress of graph adversarial defense algorithm is reviewed. First, the work background and related concepts of graph adversarial defense are introduced, and the development process of graph adversarial defense is analyzed. Then, according to different defense strategies of the defense algorithm, the algorithms are divided into four categories, including attack detection, adversarial training, robustness certification and immu-nologic defense and the strategies of each type of defense algorithm are summarized. Furthermore, the principles and implementation of defense algorithm are analyzed, and typical algorithms are compared in terms of defense strategies, target task, advantages, disadvantages and experiments. Finally, through a comprehensive and systematic analysis of the existing graph adversarial defense algorithm, the problems and developing directions of the defense algorithm are summarized to provide help for further development of graph adversarial defense.

Diverse real world systems can be abstracted as complex networks consisting of nodes and edges as functional components. Percolation theory has shown that the failure of a few of nodes could lead to the collapse of a whole network, which brings up the network dismantling problem: How to select the least number of nodes to decompose a network into disconnected components each smaller than a predefined threshold? For its NP-hardness, many heuristic approaches have been proposed to measure and rank each node according to its importance to network structural stability. However, these measures are from a uniscale viewpoint by regarding one complex network as a flatted topology. In this article, we argue that nodes' structural importance can be measured in different scales of network topologies. Built upon recent deep learning techniques, we propose a self-supervised learning based network dismantling framework (NEES), which can hierarchically merge some compact substructures to convert a network into a coarser one with fewer nodes and edges. During the merging process, we design neural models to extract essential structures and utilize self-attention mechanisms to learn nodes' importance hierarchy in each scale. Experiments on real world networks and synthetic model networks show that the proposed NEES outperforms the state-of-the-art schemes in most cases in terms of removing the least number of target nodes to dismantle a network. The dismantling effectiveness of our neural extraction framework also highlights the emerging role of multi-scale essential structures. [ABSTRACT FROM AUTHOR]

Cross-media association mining based on heterogeneous information network(HIN) has received widespread attention. However, video is described by only a few words, leading to the lack of association between visual and textual information. As a result, the heterogeneous graph is inevitably incomplete, which brings great challenges to event mining. Fortunately, topological relationships can infer correlations between similar nodes. In view of this, a novel framework of web video event mining based on attention graph structure learning is proposed to generate a new adjacency matrix, which reconstructs the association among nodes. First, a novel heterogeneous network is constructed, while each relation subgraph is produced separately. Then, in each relational subgraph, feature graphs are generated by feature similarity, which can capture potential relationships between nodes. Simultaneously, semantic graph is also created by learning semantic structures to describe complex heterogeneous interactions between node semantics. Next, these graphs are fused by channel attention to reconstruct the correlation among nodes. Finally, graph convolutional network(GCN) is applied for web video event mining. Experiments on web videos from YouTube demonstrate that our proposed method is more effective than the state-of-the-art methods with significant improvement. [ABSTRACT FROM AUTHOR]

As most great discoveries and advancements in science and technology invariably involve the cooperation of a group of researchers, effective collaboration is the key factor. Nevertheless, finding suitable scholars and researchers to work with is challenging and, mostly, time-consuming for many. A recommender who is capable of finding and recommending collaborators would prove helpful. In this work, we utilized a life science and biomedical research database, i.e., MEDLINE, to develop a collaboration recommendation system based on novel graph neural networks, i.e., GraphSAGE and Temporal Graph Network, which can capture intrinsic, complex, and changing dependencies among researchers, including temporal user–user interactions. The baseline methods based on LightGCN and gradient boosting trees were also developed in this work for comparison. Internal automatic evaluations and external evaluations through end-users' ratings were conducted, and the results revealed that our graph neural networks recommender exhibits consistently encouraging results.

By representing the embedded components and their interactions in industrial systems as nodes and edges in a graph, Graph Neural Networks (GNNs) have achieved outstanding results due to their ability to model statistical correlations. However, these correlations may not capture the true causal relationships within the data, thereby impairing the model's performance in fault diagnosis. To address this issue, an Information-based Gradient enhanced Causal Learning Graph Neural Network (IGCL-GNN) is proposed for fault diagnosis of complex industrial processes. First, the information theory in graph representations is theoretically analyzed and the optimization objectives are derived separately for the causal and non-causal parts of the graph neural network, which decouple it into a multi-objective optimization problem. Then, to optimize such problem, a causal disentanglement layer is designed in the graph network that could effectively separate causal and non-causal information in graph representations. Thirdly, a novel gradient reactivation method is proposed to dynamically filter features from the disentangled layers, thereby capture the causal representations of graph data more accurately. For robust and efficient optimization, the multi-objective gradient descent algorithm is employed in this paper. Finally, comparative experiments were conducted on the three-phase flow facility (TFF) dataset, achieving a fault diagnosis accuracy of 98.42% for our proposed method. • Explores causal feature learning in fault diagnosis from an information theory perspective. • Presents a gradient reactivation module for reliable causal subgraph extraction. • Improves the relevance of predictions to causal factors while suppressing the extraction of non-causal features. • Ensures the maximum extraction of real relevant information through the Pareto optimality of the two objectives. [ABSTRACT FROM AUTHOR]

Soil heavy metal contamination has emerged as a global environmental concern, posing significant risks to human health and ecosystem integrity. Hyperspectral technology, with its non-invasive, non-destructive, large-scale, and high spectral resolution capabilities, shows promising applications in monitoring soil heavy metal pollution. Traditional monitoring methods are often time-consuming, labor-intensive, and expensive, limiting their effectiveness for rapid, large-scale assessments. This study introduces a novel deep learning method, SpecMet, for estimating heavy metal concentrations in naturally contaminated agricultural soils using hyperspectral data. The SpecMet model extracts features from hyperspectral data using convolutional neural networks (CNNs) and achieves end-to-end prediction of soil heavy metal concentrations by integrating attention mechanisms and graph neural networks. Results demonstrate that the OR-SpecMet model, which utilizes raw spectral data, achieves optimal prediction performance, significantly surpassing traditional machine learning methods such as multiple linear regression, partial least squares regression, and support vector machine regression in estimating concentrations of lead (Pb), copper (Cu), cadmium (Cd), and mercury (Hg). Moreover, training specialized OR-SpecMet models for individual heavy metals better accommodates their unique spectral-concentration relationships, enhancing overall estimation accuracy while achieving a 20.3 % improvement in predicting low-concentration mercury. The OR-SpecMet method showcases the superior performance and extensive application potential of deep learning techniques in precise soil heavy metal pollution monitoring, offering new insights and reliable technical support for soil pollution prevention and agricultural ecosystem protection. The code and datasets used in this study are publicly available at: https://github.com/zhang2lei/metal.git. • Using attention mechanism and graph neural network for metal concentration inversion. • OR-SpecMet outperformed traditional methods in retrieving Pb, Cu, Cd, and Hg concentrations. • Metal-specific OR-SpecMet boosted prediction accuracy. [ABSTRACT FROM AUTHOR]

Network slicing has emerged as a promising networking paradigm to provide resources tailored for Industry 4.0 and diverse services in 5G networks. However, the increased network complexity poses a huge challenge in network management due to virtualized infrastructure and stringent quality-of-service requirements. Digital twin (DT) technology paves a way for achieving cost-efficient and performance-optimal management, through creating a virtual representation of slicing-enabled networks digitally to simulate its behaviors and predict the time-varying performance. In this article, a scalable DT of network slicing is developed, aiming to capture the intertwined relationships among slices and monitor the end-to-end (E2E) metrics of slices under diverse network environments. The proposed DT exploits the novel graph neural network model that can learn insights directly from slicing-enabled networks represented by non-Euclidean graph structures. Experimental results show that the DT can accurately mirror the network behaviour and predict E2E latency under various topologies and unseen environments. [ABSTRACT FROM AUTHOR]

Graph-based methods have been widely used by the document image analysis and recognition community, as the different objects and the content in document images is best represented by this powerful structural representation. Designing of novel computation tools for processing these graph-based structural representations has always remained a hot topic of research. Recently, Graph Neural Network (GNN) have been used for solving different problems in the domain of document image analysis and recognition. In this article we take forward the state of the art by presenting a new approach to gather the symbolic and numeric information from the nodes and edges of a graph. We use this information to learn a Graph Neural Network (GNN). The experimentation on the recognition of handwritten letters and graphical symbols shows that the proposed approach is an interesting contribution to the growing set of GNN-based methods for document image analysis and recognition.

Copyright of Journal of Frontiers of Computer Science & Technology is the property of Beijing Journal of Computer Engineering & Applications Journal Co Ltd. and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Federated learning has been applied to various tasks in intelligent transportation systems to protect data privacy through decentralized training schemes. The majority of the state-of-the-art models in intelligent transportation systems (ITS) are graph neural networks (GNN)-based for spatial information learning. When applying federated learning to the ITS tasks with GNN-based models, the existing frameworks can only protect the data privacy; however, ignore the one of topological information of transportation networks. In this article, we propose a novel federated learning framework to tackle this problem. Specifically, we introduce a differential privacy-based adjacency matrix preserving approach for protecting the topological information. We also propose an adjacency matrix aggregation approach to allow local GNN-based models to access the global network for a better training effect. Furthermore, we propose a GNN-based model named attention-based spatial-temporal graph neural networks (ASTGNN) for traffic speed forecasting. We integrate the proposed federated learning framework and ASTGNN as FASTGNN for traffic speed forecasting. Extensive case studies on a real-world dataset demonstrate that FASTGNN can develop accurate forecasting under the privacy preservation constraint. [ABSTRACT FROM AUTHOR]

The recent success of graph neural networks (GNNs) in the area of pattern recognition (PR) has increased the interest of researchers to use these frameworks in non-euclidean structures. This non-euclidean structure includes graphs or manifolds that are called geometric deep learning (GDL). It has opened a new direction for researchers to deal with graphs using deep learning in document processing, outperforming conventional methods. We propose a Deep Graph Neural Network (DGNN) classifier-based on additive angular margin loss for the classification task in document analysis. Another contribution of this work is to investigate the performance of a DGNN as a classifier using different loss functions, which helps to minimize the loss for the document analysis problem. We compare additive angular margin loss, Cosine angular margin loss, and multiplicative angular margin loss. Furthermore, we give a comparison between the mentioned loss functions and the Softmax loss function. We also present the comparisons of results using different graph edit distance (GED) methods. Our quantitative results suggest, that by applying the additive angular marginal loss function makes more compact intra-class ability and increases the inter-class discrepancy which enhances the discriminating power of the DGNN. Enhancing the decision boundaries between the classes increase the intra-class compactness and inter-class discrimination power of the model.

Objective. Brain-computer interface (BCI) systems with large directly accessible instruction sets are one of the difficulties in BCI research. Research to achieve high target resolution (⩾100) has not yet entered a rapid development stage, which contradicts the application requirements. Steady-state visual evoked potential (SSVEP) based BCIs have an advantage in terms of the number of targets, but the competitive mechanism between the target stimulus and its neighboring stimuli is a key challenge that prevents the target resolution from being improved significantly. Approach. In this paper, we reverse the competitive mechanism and propose a frequency spatial multiplexing method to produce more targets with limited frequencies. In the proposed paradigm, we replicated each flicker stimulus as a 2 × 2 matrix and arrange the matrices of all frequencies in a tiled fashion to form the interaction interface. With different arrangements, we designed and tested three example paradigms with different layouts. Further we designed a graph neural network that distinguishes between targets of the same frequency by recognizing the different electroencephalography (EEG) response distribution patterns evoked by each target and its neighboring targets. Main results. Extensive experiment studies employing eleven subjects have been performed to verify the validity of the proposed method. The average classification accuracies in the offline validation experiments for the three paradigms are 89.16%, 91.38%, and 87.90%, with information transfer rates (ITR) of 51.66, 53.96, and 50.55 bits/min, respectively. Significance. This study utilized the positional relationship between stimuli and did not circumvent the competing response problem. Therefore, other state-of-the-art methods focusing on enhancing the efficiency of SSVEP detection can be used as a basis for the present method to achieve very promising improvements., (Creative Commons Attribution license.)

In the Islamic domain, Hadiths hold significant importance, standing as crucial texts following the Holy Quran. Each Hadith contains three main parts: the ISNAD (chain of narrators), TARAF (starting part, often from Prophet Muhammad), and MATN (Hadith content). ISNAD, a chain of narrators involved in transmitting that particular MATN. Hadith scholars determine the trustworthiness of the transmitted MATN by the quality of the ISNAD. The ISNAD's data is available in its original Arabic language, with narrator names transliterated into English. This paper presents the Multi-IsnadSet (MIS), that has great potential to be employed by the social scientist and theologist. A multi-directed graph structure is used to represents the complex interactions among the narrators of Hadith. The MIS dataset represent directed graph which consists of 2092 nodes, representing individual narrators, and 77,797 edges represent the Sanad-Hadith connections. The MIS dataset represents multiple ISNAD of the Hadith based on the Sahih Muslim Hadith book. The dataset was carefully extracted from online multiple Hadith sources using data scraping and web crawling techniques tools, providing extensive Hadith details. Each dataset entry provides a complete view of a specific Hadith, including the original book, Hadith number, textual content (MATN), list of narrators, narrator count, sequence of narrators, and ISNAD count. In this paper, four different tools were designed and constructed for modeling and analyzing narrative network such as python library (NetworkX), powerful graph database Neo4j and two different network analysis tools named Gephi and CytoScape. The Neo4j graph database is used to represent the multi-dimensional graph related data for the ease of extraction and establishing new relationships among nodes. Researchers can use MIS to explore Hadith credibility including classification of Hadiths (Sahih=perfection in the Sanad/Dhaif=imperfection in the Sanad), and narrators (trustworthy/not). Traditionally, scholars have focused on identifying the longest and shortest Sanad between two Narrators, but in MIS, the emphasis shifts to determining the optimum/authentic Sanad, considering narrator qualities. The graph representation of the authentic and manually curated dataset will open ways for the development of computational models that could identify the significance of a chain and a narrator. The dataset allows the researchers to provide Hadith narrators and Hadith ISNAD that could be used in a wide variety of future research studies related to Hadith authentication and rules extraction. Moreover, the dataset encourages cross-disciplinary research, bridging the gap between Islamic studies, artificial intelligence (AI), social network analysis (SNA), and Graph Neural Network (GNN)., (© 2024 The Authors.)

• Proposed a novel machine learning approach based on Graph Neural Networks (GNN) for dynamic response simulation of structures subjected to dynamics loads. • The proposed GNN can simulate full-field spatiotemporal dynamics, generating response quantities, e.g., displacement, strain, and stress, at any position of structures at any time. • Extensive experiments on different structures and different types of loads demonstrated the general applicability of GNN for structural response prediction. • GNN simulations agree well the numerical model SPH in LS-DYNA, but runs an order of magnitude faster. Prediction of structural responses is essential for the analysis of structural behaviour subjected to dynamic loads. Existing approaches are limited in different ways. Experimental tests are expensive due to the requirement of intensive labour hours and specialised equipment. Numerical models can provide high-fidelity and labour-effective simulation of impact tests after proper validation, but it comes with high computational costs, which prohibit the usage of numerical methods for intensive and large-scale simulations in a design office. Data-driven machine learning approaches are also applied to structural response predictions, but they often exploit direct input–output mapping schemes that predict static field variables without capturing the dynamic response process. To close these gaps, we propose a novel machine learning approach based on graph neural networks (GNN) for full-field structural dynamics prediction. Our approach adopts a discretised representation of structures with an iterative rollout prediction scheme, and therefore it can simulate comprehensive spatiotemporal structural dynamics, providing the full potential for structural dynamics analysis. With several benchmark tests, it is demonstrated that our approach can generate accurate predictions of related field variables, e.g., displacement, strain, and stress, for different structures with a wide range of input parameters, such as structure geometry, impact speed and location. Additional interpolation and extrapolation tests are also conducted to show that our approach enjoys inherent generalisability and can produce satisfactory prediction even when all inputs are sampled outside of the training distribution. Our approach is also efficient and runs an order of magnitude faster than the commonly used numerical competitors. As the first attempt of using GNN for structural dynamics prediction and the result is promising, it is believed that GNN is well-suited for effective and efficient prediction of dynamic structural responses. [ABSTRACT FROM AUTHOR]

With the explosive growth of online information, the significant application value of recommender systems has received considerable attention. Since user–item interactions can naturally fit into graph structure data, graph neural networks (GNNs), by virtue of their strong ability in graph representation learning, have become the new state-of-the-art approach to recommender systems. Recently, GNN-based contrastive self-supervised learning (SSL) methods have received careful attention due to their superiority over graph-based recommendation under the typical supervised learning paradigm. However, to achieve state-of-the-art performance, GNN-based recommendation with SSL often needs a huge amount of negative examples and the model's performance is heavily dependent on complex data augmentations. Also, the information interaction among various augmented views is often performed under a single perspective (e.g., structure/feature space or node/graph level). In this paper, we propose a novel bootstrapped graph representation learning with local and global regularization for recommendation, i.e., BLoG, which constructs positive/negative pairs based on the aggregated node features by referring to two alternate views of the original user–item graph structure. In particular, BLoG learns user–item representations by encoding two augmented versions of a user–item bipartite graph using two separate encoders: an online encoder and a target encoder. To facilitate the information interaction between these two distinct graph encoders, we introduce local and global regularization for recommendation, where a graph structural contrastive loss and a node-level semantic loss are defined for local regularization while a graph-level contrastive loss is used for global regularization. An alternative optimization approach is used to train the online encoder and the target encoder. Experimental studies on three benchmark datasets demonstrate that BLoG achieves better recommendation accuracy than several existing baselines. • A novel GNN-based contrastive learning paradigm for recommendation is proposed. • Bootstrapped graph representation learning with local and global regularization is developed. • An alternative optimization approach is used to train the online encoder and the target encoder. • Our proposed BLoG framework outperforms several baselines. [ABSTRACT FROM AUTHOR]

Drug repurposing has gained the attention of many in the recent years. The practice of repurposing existing drugs for new therapeutic uses helps to simplify the drug discovery process, which in turn reduces the costs and risks that are associated with de novo development. Representing biomedical data in the form of a graph is a simple and effective method to depict the underlying structure of the information. Using deep neural networks in combination with this data represents a promising approach to address drug repurposing. This paper presents BEHOR a more comprehensive version of the REDIRECTION model, which was previously presented. Both versions utilize the DISNET biomedical graph as the primary source of information, providing the model with extensive and intricate data to tackle the drug repurposing challenge. This new version's results for the reported metrics in the RepoDB test are 0.9604 for AUROC and 0.9518 for AUPRC. Additionally, a discussion is provided regarding some of the novel predictions to demonstrate the reliability of the model. The authors believe that BEHOR holds promise for generating drug repurposing hypotheses and could greatly benefit the field. • A complex set of biomedical information from different sources has been integrated. • A large biomedical graph with 5 node and 12 edge types was constructed. • Graph Neural Networks were used to predict links between drugs and diseases. • Five of the top novel predictions by the model and their plausibility were studied. • Model achieves 0.9604 AUROC and 0.9518 AUPRC for the presented secondary test. [ABSTRACT FROM AUTHOR]

Trust prediction is essential to enhancing reliability and reducing risk from the unreliable node, especially for online applications in open network environments. An essential fact in trust prediction is to measure the relation of both the interacting entities accurately. However, most of the existing methods infer the trust relation between interacting entities usually rely on modeling the similarity between nodes on a graph and ignore semantic relation and the influence of negative links (e.g., distrust relation). In this paper, we proposed a relation representation learning via signed graph mutual information maximization (called SGMIM). In SGMIM, we incorporate a translation model and positive point-wise mutual information to enhance the relation representations and adopt Mutual Information Maximization to align the entity and relation semantic spaces. Moreover, we further develop a sign prediction model for making accurate trust predictions. We conduct link sign prediction in trust networks based on learned the relation representation. Extensive experimental results in four real-world datasets on trust prediction task show that SGMIM significantly outperforms state-of-the-art baseline methods.

• A systematic comparison of feature types for IL conductivity prediction using ML. • Structural and/or molecular features used to train the ML model, using a dataset of 2,684 ILs. • Obtained MAE, RMSE, and R2 of 0.470, 0.677 and 0.937, respectively. Predicting relevant ionic liquid (IL) properties like ionic conductivity using machine learning techniques has attracted significant interest in literature due to the diverse IL design space. The accuracy of these predictions depends on the types of features used to train the machine learning-based models. Previous works using machine learning to predict IL ionic conductivity were limited in dataset size and diversity of ILs. Using graphical neural network and a multi-layer perceptron on a significantly more extensive and diverse dataset, we predict the ionic conductivity of ionic liquids. We establish a systematic comparison by evaluating three feature sets: structural features learned through graph neural networks, molecular features, and a combination of structural and molecular features. The better conductivity prediction performance observed herein using structural features only (compared to molecular features only) reinforces the importance of chemical structure in characterising the ionic conductivity of ILs. The best prediction results were obtained with the model that combined structural and molecular features. Mean absolute error, root mean squared error, and coefficient of determination of 0.470, 0.677 and 0.937, respectively, were obtained by the model with the combined feature sets, using a dataset of 2,684 ILs. A well-informed choice of what molecular features to combine with structural features is critical for better property prediction accuracy. [ABSTRACT FROM AUTHOR]

As most great discoveries and advancements in science and technology invariably involve the cooperation of a group of researchers, effective collaboration is the key factor. Nevertheless, finding suitable scholars and researchers to work with is challenging and, mostly, time-consuming for many. A recommender who is capable of finding and recommending collaborators would prove helpful. In this work, we utilized a life science and biomedical research database, i.e., MEDLINE, to develop a collaboration recommendation system based on novel graph neural networks, i.e., GraphSAGE and Temporal Graph Network, which can capture intrinsic, complex, and changing dependencies among researchers, including temporal user-user interactions. The baseline methods based on LightGCN and gradient boosting trees were also developed in this work for comparison. Internal automatic evaluations and external evaluations through end-users' ratings were conducted, and the results revealed that our graph neural networks recommender exhibits consistently encouraging results., Competing Interests: The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest., (Copyright © 2022 Zhu and Yaseen.)