Graph Convolutional Network Research Articles

Combine multi-order representation learning and frame optimization learning for skeleton-based action recognition

Skeleton-based action recognition has broad application prospects in many fields such as virtual reality. Currently, the most popular way is to employ Graph Convolutional Networks (GCNs) or Hypergraph Convolutional Networks (HGCNs) for this task. However, GCN-based methods may heavily rely on the physical connectivity relationship between joints while lack the capture of higher-order information about interactions among distant joints, and HGCN-based methods usually introduce unnecessary noise when capturing low-order information of skeleton structures with simple topology. Besides, the current methods do not deal well with redundant frames and confusing frames. These limitations hinder the improvement of recognition accuracy. In this paper, we propose a novel network, called Hyper-Net, which combines multi-order representation learning and frame optimization learning for skeleton-based action recognition. Specifically, the proposed Hyper-Net contains Temporal-Channel Aggregation Graph Convolution (TCA-GC), Spatial-Temporal Aggregation Hypergraph Convolution (STA-HC) and Frame Optimization Learning (F-OL) modules. The TCA-GC aggregates low-order and local information from simple joint and bone topologies across different temporal and channel dimensions. The STA-HC captures high-order and global information from complex motion streams as well as solving the problem of spatial-temporal weight imbalance. The F-OL can adaptively extract key frames and distinguish confusing frames, thus improving the ability of the network to recognize confusing actions. A large number of experiments are conducted on the NTU RGB+D, NTU RGB+D 120 and NW-UCLA datasets for action recognition task. Experimental results demonstrate the superiority and effectiveness of the proposed network.

User preference and social relationship-aware recommendations base on a novel light graph convolutional network

User preference and social relationship-aware recommendations base on a novel light graph convolutional network

SWL-LSE: A Dataset of Health-Related Signs in Spanish Sign Language with an ISLR Baseline Method

Progress in automatic sign language recognition and translation has been hindered by the scarcity of datasets available for the training of machine learning algorithms, a challenge that is even more acute for languages with smaller signing communities, such as Spanish. In this paper, we introduce a dataset of 300 isolated signs in Spanish Sign Language, collected online via a web application with contributions from 124 participants, resulting in a total of 8000 instances. This dataset, which is openly available, includes keypoints extracted using MediaPipe Holistic. The goal of this paper is to describe the construction and characteristics of the dataset and to provide a baseline classification method using a spatial–temporal graph convolutional network (ST-GCN) model, encouraging the scientific community to improve upon it. The experimental section offers a comparative analysis of the method’s performance on the new dataset, as well as on two other well-known datasets. The dataset, code, and web app used for data collection are freely available, and the web app can also be used to test classifier performance on-line in real-time.

SMAR: self-supervised mobile application recommendation based on graph convolutional networks

PurposeWith the exponential growth of mobile applications, recommending suitable mobile applications to users becomes a critical challenge. Although existing methods have made achievements in mobile application recommendation by leveraging graph convolutional networks (GCNs), they suffer from two limitations: the reliance on a singular acquisition path leads to signal sparsity, and the neighborhood aggregation method exacerbates the adverse impact of noisy interactions. This paper aims to propose SMAR, a self-supervised mobile application recommendation approach based on GCN, which is designed to overcome existing challenges by using self-supervised learning to create an auxiliary task.Design/methodology/approachIn detail, this method uses three distinct data augmentation techniques node dropout, edge dropout and random walk, which create varied perspectives of each node. Then compares these perspectives, aiming to ensure uniformity across different views of the same node while maintaining the differences between separate nodes. Ultimately, auxiliary task is combined with the primary supervised task using a multi-task learning framework, thereby refining the overall mobile application recommendation process.FindingsExtensive experiments on two real datasets demonstrate that SMAR achieves better Recall and NDCG performances than other strong baselines, validating the effectiveness of the proposed method.Originality/valueIn this paper, the authors introduce self-supervised learning into mobile application recommendation approach based on GCNs. This method enhances traditional supervised tasks by using auxiliary task to provide additional information, thereby improving signal accuracy and reducing the influence of noisy interactions in mobile application recommendations.

A multi-modality and multi-granularity collaborative learning framework for identifying spatial domains and spatially variable genes.

Recent advances in spatial transcriptomics technologies have provided multi-modality data integrating gene expression, spatial context, and histological images. Accurately identifying spatial domains and spatially variable genes is crucial for understanding tissue structures and biological functions. However, effectively combining multi-modality data to identify spatial domains and determining SVGs closely related to these spatial domains remains a challenge. In this study, we propose spatial transcriptomics Multi-modality and Multi-granularity Collaborative Learning (spaMMCL). For detecting spatial domains, spaMMCL mitigates the adverse effects of modality bias by masking portions of gene expression data, integrates gene and image features using a shared graph convolutional network, and employs graph self-supervised learning to deal with noise from feature fusion. Simultaneously, based on the identified spatial domains, spaMMCL integrates various strategies to detect potential SVGs at different granularities, enhancing their reliability and biological significance. Experimental results demonstrate that spaMMCL substantially improves the identification of spatial domains and SVGs. The code and data of spaMMCL are available on Github: Https://github.com/liangxiao-cs/spaMMCL. Supplementary data are available at Bioinformatics online.

A traffic speed prediction algorithm for dynamic spatio-temporal graph convolutional networks based on attention mechanism

A traffic speed prediction algorithm for dynamic spatio-temporal graph convolutional networks based on attention mechanism

Spotting Leaders in Organizations with Graph Convolutional Networks, Explainable Artificial Intelligence, and Automated Machine Learning

Over the past few decades, the study of leadership theory has expanded across various disciplines, delving into the intricacies of human behavior and defining the roles of individuals within organizations. Its primary objective is to identify leaders who play significant roles in the communication flow. In addition, behavioral theory posits that leaders can be distinguished based on their daily conduct, while social network analysis provides valuable insights into behavioral patterns. Our study investigates five and six types of social networks frequently observed in different organizations. This study is conducted using datasets we collected from an IT company and public datasets collected from a manufacturing company for the thorough evaluation of prediction performance. We leverage PageRank and effective word embedding techniques to obtain novel features. State-of-the-art performance is obtained using various statistical machine learning methods, graph convolutional networks (GCN), automated machine learning (AutoML), and explainable artificial intelligence (XAI). More specifically, our approach can achieve state-of-the-art performance with an accuracy close to 90% for leaders identification with data from projects of different types. This investigation contributes to the establishment of sustainable leadership practices by aiding organizations in retaining their leadership talent.

Neural network-augmented differentiable finite element method for boundary value problems

Classical numerical methods such as finite element method (FEM) face limitations due to their low efficiency when addressing large-scale problems. As a novel paradigm, the physics-informed neural network (PINN) has demonstrated significant potential to solve partial differential equations. However, conventional PINNs utilize meshless control at discrete sampling points, which limits their ability to effectively handle complex boundaries. Moreover, catastrophic failure may occur in the deep energy method (DEM, a specific type of PINN). To handle these challenges, this study proposes a Neural Network-augmented Differentiable Finite Element Method (NNDFEM) by combining PINN and finite element approximation. In NNDFEM, the neural network backend solely predicts nodal variables. Derivatives and complex boundary conditions can be well handled by the finite element frontend. The governing equation over the domain, Dirichlet, and Neumann boundary conditions are directly enforced on the finite element frontend. Thus, losses of boundary conditions in PINN are rendered unnecessary. The overfitting problem in DEM is also significantly mitigated. Fully connected neural network (FCNN), modified FCNN, and graph-convolutional network are tested as backends. NNDFEM circumvents nodal force calculation and matrix assembly in FEM. Functional losses of linear elasticity, finite strain nonlinear elasticity, heat conduction, and flow in porous media are validated. A systematic exploration unveils the role of 3D finite element mesh. For large-scale problems, a multi-fidelity learning strategy is employed. Thus, the three-dimensional case with over three million degrees of freedom trains well in two minutes. Benefiting from the fast inference of the neural network backend, the forward pass is 8,550 times faster than FEM.

A bike-sharing demand prediction model based on Spatio-Temporal Graph Convolutional Networks

Shared bikes, as an eco-friendly transport mode, facilitate short commutes for urban dwellers and help alleviate traffic. However, the prevalent station-based strategy for bike placements often overlooks urban zones, cycling patterns, and more, resulting in underutilized bikes. To address this, we introduce the Spatio-Temporal Bike-sharing Demand Prediction (ST-BDP) model, leveraging multi-source data and Spatio-Temporal Graph Convolutional Networks (STGCN). This model predicts spatial user demand for bikes between stations by constructing a spatial demand graph, accounting for geographical influences. For precision, ST-BDP integrates an attention-based graph convolutional network for station demand graph’s temporal-spatial features, and a sequential convolutional network for multi-source data (e.g., weather, time). In real dataset, experimental results show that ST-BDP has excellent performance with mean absolute error (MAE) = 1.62, mean absolute percentage error (MAPE) = 15.82%, symmetric mean absolute percentage error (SMAPE) = 16.14%, and root mean square error (RMSE) = 2.36, outperforming the baseline techniques. This highlights its predictive accuracy and potential to guide future bike-sharing policies.

Deep and shallow feature fusion framework for remote sensing open pit coal mine scene recognition

Understanding land use and damage in open-pit coal mining areas is crucial for effective scientific oversight and management. Current recognition methods exhibit limitations: traditional approaches depend on manually designed features, which offer limited expressiveness, whereas deep learning techniques are heavily reliant on sample data. In order to overcome the aforementioned limitations, a three-branch feature extraction framework was proposed in the present study. The proposed framework effectively fuses deep features (DF) and shallow features (SF), and can accomplish scene recognition tasks with high accuracy and fewer samples. Deep features are enhanced through a neighbouring feature attention module and a Graph Convolutional Network (GCN) module, which capture both neighbouring features and the correlation between local scene information. Shallow features are extracted using the Gray-Level Co-occurrence Matrix (GLCM) and Gabor filters, which respectively capture local and overall texture variations. Evaluation results on the AID and RSSCN7 datasets demonstrate that the proposed deep feature extraction model achieved classification accuracies of 97.53% and 96.73%, respectively, indicating superior performance in deep feature extraction tasks. Finally, the two kinds of features were fused and input into the particle swarm algorithm optimised support vector machine (PSO-SVM) to classify the scenes of remote sensing images, and the classification accuracy reached 92.78%, outperforming four other classification methods.

DGSIST: Clustering spatial transcriptome data based on deep graph structure Infomax

Although spatial transcriptomics data provide valuable insights into gene expression profiles and the spatial structure of tissues, most studies rely solely on gene expression information, underutilizing the spatial data. To fully leverage the potential of spatial transcriptomics and graph neural networks, the DGSI (Deep Graph Structure Infomax) model is proposed. This innovative graph data processing model uses graph convolutional neural networks and employs an unsupervised learning approach. It maximizes the mutual information between graph-level and node-level representations, emphasizing flexible sampling and aggregation of nodes and their neighbors. This effectively captures and incorporates local information from nodes into the overall graph structure. Additionally, this paper developed the DGSIST framework, an unsupervised cell clustering method that integrates the DGSI model, SVD dimensionality reduction algorithm, and k-means++ clustering algorithm. This aims to identify cell types accurately. DGSIST fully uses spatial transcriptomics data and outperforms existing methods in accuracy. Demonstrations of DGSIST’s capability across various tissue types and technological platforms have shown its effectiveness in accurately identifying spatial domains in multiple tissue sections. Compared to other spatial clustering methods, DGSIST excels in cell clustering and effectively eliminates batch effects without needing batch correction. DGSIST excels in spatial clustering analysis, spatial variation identification, and differential gene expression detection and directly applies to graph analysis tasks, such as node classification, link prediction, or graph clustering. Anticipation lies in the contribution of the DGSIST framework to a deeper understanding of the spatial organizational structures of diseases such as cancer.

Deep spatial-temporal information fusion dynamic graph convolutional network for traffic flow prediction

Abstract Predicting traffic flow is vital for advancing intelligent transportation systems, but achieving precise forecasts remains challenging. To deeply explore the complex and diverse dynamic spatio-temporal correlation of traffic data, this study proposes a multi-information fusion spatio-temporal dynamic graph convolution model for traffic flow prediction. In the model construction, firstly, a dynamic graph generation module is designed to fully capture the complex spatial correlation of the traffic network and model the changing spatio-temporal interaction; secondly, the spatial self-attention mechanism is used to dynamically focus on the spatial relationship between different nodes, and combined with the dynamic graph convolution network to extract the spatial dynamic correlation features of the road network to study deep spatiotemporal information; Finally, existing models rarely consider other traffic information related to traffic flow, and a spatiotemporal information fusion module is designed to enhance the ability of the model to perceive information, thereby improving the performance of traffic flow prediction models. Experiments are conducted on four real traffic datasets using three performance evaluation metrics and 18 baseline models. The results show that the model has higher prediction accuracy.

Trustworthiness-Driven Graph Convolutional Networks for Signed Network Embedding

The problem of representing nodes in a signed network as low-dimensional vectors, known as signed network embedding (SNE), has garnered considerable attention in recent years. While several SNE methods based on graph convolutional networks (GCNs) have been proposed for this problem, we point out that they significantly rely on the assumption that the decades-old balance theory always holds in the real-world. To address this limitation, we propose a novel GCN-based SNE approach, named as TrustSGCN, which corrects for incorrect embedding propagation in GCN by utilizing the trustworthiness on edge signs for high-order relationships inferred by the balance theory. The proposed approach consists of three modules: (M1) generation of each node’s extended ego-network; (M2) measurement of trustworthiness on edge signs; and (M3) trustworthiness-aware propagation of embeddings. Specifically, TrustSGCN leverages topological information to measure trustworthiness on edge sign for high-order relationships inferred by balance theory. It then considers structural properties inherent to an input network, such as the ratio of triads, to correct for incorrect embedding propagation. Furthermore, TrustSGCN learns the node embeddings by leveraging two well-known social theories, i.e., balance and status, to jointly preserve the edge sign and direction between nodes connected by existing edges in the embedding space. The experiments on six real-world signed network datasets demonstrate that TrustSGCN consistently outperforms six state-of-the-art GCN-based SNE methods. The code is available at https://github.com/kmj0792/TrustSGCN .

A Brain Network Analysis Model for Motion Sickness in Electric Vehicles Based on EEG and fNIRS Signal Fusion

Motion sickness is a common issue in electric vehicles, significantly impacting passenger comfort. This study aims to develop a functional brain network analysis model by integrating electroencephalography (EEG) and functional near-infrared spectroscopy (fNIRS) signals to evaluate motion sickness symptoms. During real-world testing with the Feifan F7 series of new energy-electric vehicles from SAIC Motor Corp, data were collected from 32 participants. The EEG signals were divided into four frequency bands: delta-range, theta-range, alpha-range, and beta-range, and brain oxygenation variation was calculated from the fNIRS signals. Functional connectivity between brain regions was measured to construct functional brain network models for motion sickness analysis. A motion sickness detection model was developed using a graph convolutional network (GCN) to integrate EEG and fNIRS data. Our results show significant differences in brain functional connectivity between participants in motion and non-motion sickness states. The model that combined fNIRS data with high-frequency EEG signals achieved the best performance, improving the F1 score by 11.4% compared to using EEG data alone and by 8.2% compared to using fNIRS data alone. These results highlight the effectiveness of integrating EEG and fNIRS signals using GCN for motion sickness detection. They demonstrate the model’s superiority over single-modality approaches, showcasing its potential for real-world applications in electric vehicles.

Transient Stability Assessment in Power Systems: A Spatiotemporal Graph Convolutional Network Approach with Graph Simplification

Accurate and fast transient stability assessment (TSA) of power systems is crucial for safe operation. However, deep learning-based methods require long training and fail to simultaneously extract the spatiotemporal characteristics of the transient process in power systems, limiting their performance in prediction. This paper proposes a novel TSA method based on a spatiotemporal graph convolutional network with graph simplification. First, based on the topology and node information entropy of power grids, as well as the power flow of each node, the input characteristic matrix is compressed to accelerate evaluation. Then, a high-performance TSA model combining a graph convolutional network and a Gated Convolutional Network is constructed to extract the spatial features of the power grid and the temporal features of the transient process. This model establishes a mapping relationship between spatiotemporal features and their transient stability. Finally, the focal loss function has been improved to dynamically adjust the influence of samples with different levels of difficulty on model training, adaptively addressing the challenge of sample imbalance. This improvement reduces misclassification rates and enhances overall accuracy. Case studies on the IEEE 39-bus system demonstrate that the proposed method is rapid, reliable, and generalizable.

Machine learning assisted search for Fe–Co–C ternary compounds with high magnetic anisotropy

We employ a machine learning (ML)-guided framework to explore rare earth free magnetic materials, specifically focusing on Fe–Co–C ternary compounds for potential use in permanent magnets. Utilizing a specifically trained crystal graph convolutional neural network model, we efficiently screen a vast space of nearly a million substitutional structures to select 620 promising structures for further investigation by first-principles calculation. We predict five low-energy metastable Fe–Co–C compounds with formation energy less than 150 meV/atom above the convex hull. These compounds exhibit high magnetization (Js > 1.0 T) and significant magnetic anisotropy (K1 > 1.0 MJ/m3), making them promising candidates for permanent magnet applications. The phonon calculations indicate these compounds are dynamically stable. Our ML-guided framework demonstrates the utility of rapidly identifying novel materials with tailored magnetic properties.

PTNS: patent citation trajectory prediction based on temporal network snapshots

With the rapid development of science and technology, the pace of development in the knowledge economy is accelerating. Intellectual property, especially patents, is a strategic resource for technological innovation and a crucial support for building an innovative country. Therefore, it is particularly important to predict patents with high value and strong impact from the numerous and uneven-quality patents. However, patent citation behavior involves many uncertainties, and it is difficult to capture its temporal variations effectively. Therefore, this paper proposes a patent citation trajectory prediction model (PTNS) based on temporal network snapshots. It adopts relational graph convolutional networks (R-GCN) to learn the complex relationships among multiple attributes of patents and utilizes bidirectional long short-term memory networks (BiLSTM) to aggregate the temporal evolution differences of patents. Subsequently, principal component analysis (PCA) is used to explore the evolution characteristics of patent citations in depth, thereby capturing the aging effect and the ‘sleeping beauty’ phenomenon. Compared with other baselines, the PTNS performs well. In predicting new, grown, and random patents, the RMSLE decreases by approximately 0.04, 0.14, and 0.18 respectively, while the MALE decreases by approximately 0.04, 0.12, and 0.16 respectively.

Graph Neural Networks: A Bibliometric Mapping of the Research Landscape and Applications

Graph neural networks (GNNs) are deep learning algorithms that process graph-structured data and are suitable for applications such as social networks, physical models, financial markets, and molecular predictions. Bibliometrics, a tool for tracking research evolution, identifying milestones, and assessing current research, can help identify emerging trends. This study aims to map GNN applications, research directions, and key contributors. An analysis of 40,741 GNN-related publications from the Web Science Core Collection reveals a rising trend in GNN publications, especially since 2018. Computer Science, Engineering, and Telecommunications play significant roles in GNN research, with a focus on deep learning, graph convolutional networks, neural networks, convolutional neural networks, and machine learning. China and the USA combined account for 76.4% of the publications. Chinese universities concentrate on graph convolutional networks, deep learning, feature extraction, and task analysis, whereas American universities focus on machine learning and deep learning. The study also highlights the importance of Chemistry, Physics, Mathematics, Imaging Science & Photographic Technology, and Computer Science in their respective knowledge communities. In conclusion, the bibliometric analysis provides an overview of GNN research, showing growing interest and applications across various disciplines, and highlighting the potential of GNNs in solving complex problems and the need for continued research and collaboration.

Adaptive fusion graph convolutional network based interpretable fault diagnosis method for HVAC systems enhanced by unlabeled data

Fault diagnosis is critical in maintaining the operational stability and reliability of Heating, Ventilation, and Air Conditioning (HVAC) systems, which are crucial for ensuring indoor environmental quality and energy efficiency in buildings. However, traditional fault diagnosis methodologies face substantial challenges in accurately capturing the dynamic interactions within these systems and effectively utilizing unlabeled data. To overcome these limitations, an innovative fault diagnosis approach utilizing the Adaptive Fusion Graph Convolution Network (AFGCN) is proposed in this paper. This method significantly enhances the model's learning and inference abilities, particularly in scenarios with limited labeled data, by adaptively integrating the associative graph features of both unlabeled and labeled data. Furthermore, to augment the transparency and trustworthiness of the diagnostic outcomes, this paper introduces an interpretability analysis module. This module is designed to quantify the contribution of each sensor node in the fault diagnosis process. Experimental evaluations using the ASHRAE RP-1043, actual building operating chillers datasets, and LBNL FDD Data Sets_SDAHU indicate that this method offers substantial performance improvements in diagnosing faults in HVAC systems.

Predicting protein complexes in protein interaction networks using Mapper and graph convolution networks

Protein complexes are groups of interacting proteins that are central to multiple biological processes. Studying protein complexes can enhance our understanding of cellular functions and malfunctions and thus support the development of effective disease treatments. High-throughput experimental techniques allow the generation of large-scale protein-protein interaction datasets. Accordingly, various computational approaches to predict protein complexes from protein-protein interactions were presented in the literature. They are typically based on networks in which nodes and edges represent proteins and their interactions, respectively. State-of-the-art approaches mainly rely on clustering static networks to identify complexes. However, since protein interactions are highly dynamic in nature, recent approaches seek to model such dynamics by typically integrating gene expression data and identifying protein complexes accordingly. We propose MComplex, a method that utilizes time-series gene expression with interaction data to generate a temporal network which is passed to a generative adversarial network whose generator is a graph convolutional network. This creates embeddings which are then analyzed using a modified graph-based version of the Mapper algorithm to predict corresponding protein complexes. We test our approach on multiple benchmark datasets and compare identified complexes against gold-standard protein complex datasets. Our results show that MComplex outperforms existing methods in several evaluation aspects, namely recall and maximum matching ratio as well as a composite score covering aggregated evaluation measures. The code and data are available for free download from https://github.com/LeonardoDaou/MComplex.