Graph Convolutional Network Model Research Articles

SARW: Similarity-Aware Random Walk for GCN

Graph Convolutional Network (GCN) is an important method for learning graph representations of nodes. For large-scale graphs, the GCN could meet with the neighborhood expansion phenomenon, which makes the model complexity high and the training time long. An efficient solution is to adopt graph sampling techniques, such as node sampling and random walk sampling. However, the existing sampling methods still suffer from aggregating too many neighbor nodes and ignoring node feature information. Therefore, in this paper, we propose a new subgraph sampling method, namely, Similarity-Aware Random Walk (SARW), for GCN with large-scale graphs. A novel similarity index between two adjacent nodes is proposed, describing the relationship of nodes with their neighbors. Then, we design a sampling probability expression between adjacent nodes using node feature information, degree information, neighbor set information, etc. Moreover, we prove the unbiasedness of the SARW-based GCN model for node representations. The simplified version of SARW (SSARW) has a much smaller variance, which indicates the effectiveness of our subgraph sampling method in large-scale graphs for GCN learning. Experiments on six datasets show our method achieves superior performance over the state-of-the-art graph sampling approaches for the large-scale graph node classification task.

Search-engine-based surveillance using artificial intelligence for early detection of coronavirus disease outbreak

The search-engine-based surveillance methods for the early warning and prediction of infectious diseases cannot achieve search engine keywords automatic filtering and real-time updating, lead to powerless for the early warning of emerging infectious diseases. The aim of this study is to develop an artificial intelligence (AI) method for search-engine-based surveillance to improve the early warning ability for emerging infectious diseases. The 32 keywords (444 million search queries) that may be related to the coronavirus disease (COVID-19) outbreak was collected from December 18, 2019 to February 11, 2020 from Baidu’s search engine database. The graph convolution network (GCN) model was used to select search engine keywords automatically, and then, multiple linear regression was performed to explore the relationship between the daily query frequencies of keywords and daily new cases. The GCN model was used to automatically select keywords. The prediction trend of the GCN model was highly consistent with the true curve with a mean absolute error of 81.65. Three keywords including “epidemic”, “mask” and “coronavirus” were selected. The selection keywords in the search queries were highly correlated with the daily number of confirmed cases (r = 0.96, 0.94, and 0.89; P < 0.01). An abnormal initial peak (3.05 times the normal volume) in queries appeared on December 31, 2019, which could have served as an early warning signal for an outbreak. Of particular concern, 17.5% of query volume originated from the Hubei Province, 51.15% of which was from Wuhan City. The coefficients of determination (R2) of our constructed model were 0.88, 0.88, 0.84, 0.77, 0.77, 0.75, 0.73, and 0.73 for a time lag of 0–7 days, respectively, using selection keywords. The model we constructed was used in the Beijing Xinfadi outbreak as an independent test dataset, which successfully predicted the daily numbers of cases for the following days and detected an early signal during the Beijing Xinfadi outbreak (R2 = 0.79). In this paper search-engine-based surveillance based on the AI method was established for the early detection of the COVID-19 epidemic for the first time. The model achieves automatic filtering and real-time updating of search engine keywords and can effectively detect the early signals of emerging infectious diseases.

A community detection model using node embedding approach and graph convolutional network with clustering technique

The communities in a network have distinct characteristics and interrelationships. Community discovery methods based on node embedding and deep learning have surpassed spectral clustering and statistical inference as the preferred methods for handling high-dimensional network data. Community detection in graph networks using graph neural networks (GNNs) is a rapidly expanding field of study that has attracted much attention recently. Inspired by the graph convolutional network (GCN) approach, promising results in evaluating graph structure data, a different method for discovering communities using node embedding using Node2Vec via graph convolutional networks (CD2NE-GCN) model has been proposed. It has been evaluated against the most recent standard methods of graph attention network (GAT), GCN-based approach for Unsupervised Community Detection (GUCD), and Graph Adversarial Networks (GAN). In this paper, a technique for community detection that combines the ideas of graph neural network design with node embedding via message forwarding has been presented. The experimentation has made use of five benchmark datasets: Karate, Cora, CiteSeer, Facebook, and DBLP. In the first set of trials, the node2vec uses the skip-gram with negative sampling (SGNS) approach via biased random walks of the second order to produce node embeddings and feed them to the graph convolutional network. The weighted average matrix and scaled normalized matrices of low-dimensional vertices have been calculated. In the next set of experiments, the average weighted matrix was sent to the 3-layer GCN with the activation function for generating feature vectors and applied clustering technique for predicting communities. The CD2NE-GCN approach outperforms previous literature by detecting graph network communities with accuracy scores of 98.45%, 99.04%, 96.39%, 98.79%, and 97.67% for Karate, Cora, CiteSeer, Facebook, and DBLP datasets, respectively. Thus, graph convolutional network models with a node embedding approach enhance the accuracy.

Cloud-Graph: A feature interaction graph convolutional network for remote sensing image cloud detection

Convolutional neural networks (CNNs) have made significant progress in the field of cloud detection in remote sensing images thanks to their powerful feature representation capabilities. Existing methods typically aggregate low-level features containing details and high-level features containing semantics to make full use of both features to accurately detect cloud regions. However, CNNs are still limited in their ability to reason about the relationships between features, while not being able to model context well. To overcome this problem, this paper designs a novel feature interaction graph convolutional network model that extends the feature fusion process of convolutional neural networks from Euclidean space to non-Euclidean space. The algorithm consists of three main components: remote sensing image feature extraction, feature interaction graph reasoning, and high-resolution feature recovery. The algorithm constructs a feature interaction graph reasoning (FIGR) module to fully interact with low-level and high-level features and then uses a residual graph convolutional network to infer feature higher-order relationships. The network model effectively alleviates the problem of a semantic divide in the feature fusion process, allowing the aggregated features to fuse valuable details and semantic information. The algorithm is designed to better detect clouds with complex cloud layers in remote sensing images with complex cloud shape, size, thickness, and cloud-snow coexistence. Validated on publicly available 38-Cloud and SPARCS datasets and the paper’s own Landsat-8 cloud detection dataset with higher spatial resolution, the proposed method achieves competitive performance under different evaluation metrics. Code is available at https://github.com/HaiLei-Fly/CloudGraph.

MedGCN: An IoT-edge thrombus graph convolutional network for accurate prediction and prescription diagnosis of vascular occlusive diseases from unstructured clinical reports

The timely and accurate prediction and diagnosis of vascular occlusive diseases are pivotal in enhancing patient outcomes. This research addresses this critical healthcare challenge by introducing MedGCN, a novel thrombus graph convolutional network model. MedGCN is specifically designed for the precise prediction of prescriptions and diagnoses based on unstructured clinical diagnostic reports. The model synergistically blends OpenAI’s GPT4 and a uniquely designed Cross Fusion Graph Convolutional Network (CF-GCN) to ensure the meticulous fusion of knowledge. We delve into nine distinct learning tasks, encompassing both prescription and diagnosis and employ a multi-label classification GCN pretraining technique to assess them. Our evaluation underscores MedGCN’s robust predictive prowess across various tasks. By amalgamating cutting-edge AI paradigms with IoT edge computing, this research not only bolsters the efficacy of medical decision-making but also champions patient privacy. The methodologies and findings presented herein hold immense potential for deployment in IoT frameworks, thus proffering swift and precise assistance in medical decision-making and addressing a paramount need in the contemporary healthcare landscape.

Prediction of Membrane Protein Amphiphilic Helix Based on Horizontal Visibility Graph and Graph Convolution Network.

Membrane protein amphiphilic helices play an important role in many biological processes. Based on the graph convolution network and the horizontal visibility graph the prediction method of membrane protein amphiphilic helix structure is proposed in this paper. The new dataset of amphiphilic helix is constructed. In this paper, we propose the novel feature extraction method, which characterize the amphiphilicity of membrane protein. We also extract three commonly used protein features together with the new features as protein node features. The neighbor information and long-distance dependence information of proteins are further extracted by sliding window and bidirectional long-short term memory network respectively. From the perspective of horizontal visibility algorithm, we transform protein sequences into complex networks to obtain the graph features of proteins. Then, graph convolutional network model is employed to predict the amphiphilic helix structure of membrane protein. A rigorous ten-fold cross-validation shows that the proposed method outperforms other AH prediction methods on the newly constructed dataset.

Predicting miRNA-Disease Associations From miRNA-Gene-Disease Heterogeneous Network With Multi-Relational Graph Convolutional Network Model.

MiRNAs are reported to be linked to the pathogenesis of human complex diseases. Disease-related miRNAs may serve as novel bio-marks and drug targets. This work focuses on designing a multi-relational Graph Convolutional Network model to predict miRNA-disease associations (HGCNMDA) from a Heterogeneous network. HGCNMDA introduces a gene layer to construct a miRNA-gene-disease heterogeneous network. We refine the features of nodes into initial and inductive features so that the direct and indirect associations between diseases and miRNA can be considered simultaneously. Then HGCNMDA learns feature embeddings for miRNAs and disease through a multi-relational graph convolutional network model that can assign appropriate weights to different types of edges in the heterogeneous network. Finally, the miRNA-disease associations were decoded by the inner product between miRNA and disease feature embeddings. We apply our model to predict human miRNA-disease associations. The HGCNMDA is superior to the other state-of-the-art models in identifying missing miRNA-disease associations and also performs well on recommending related miRNAs/diseases to new diseases/ miRNAs. The codes are available at https://github.com/weiba/HGCNMDA.

Action Recognition Based on GCN with Adjacency Matrix Generation Module and Time Domain Attention Mechanism

Different from other computer vision tasks, action recognition needs to process larger-scale video data. How to extract and analyze the effective parts from a huge amount of video information is the main difficulty of action recognition technology. In recent years, due to the outstanding performance of Graph Convolutional Networks (GCN) in many fields, a new solution to the action recognition algorithm has emerged. However, in current GCN models, the constant physical adjacency matrix makes it difficult to mine synergistic relationships between key points that are not directly connected in physical space. Additionally, a simple time connection of skeleton data from different frames makes each frame in the video contribute equally to the recognition results, which increases the difficulty of distinguishing action stages. In this paper, the information extraction ability of the model has been optimized in the space domain and time domain, respectively. In the space domain, an Adjacency Matrix Generation (AMG) module, which can pre-analyze node sets and generate an adaptive adjacency matrix, has been proposed. The adaptive adjacency matrix can help the graph convolution model to extract the synergistic information between the key points that are crucial for recognition. In the time domain, the Time Domain Attention (TDA) mechanism has been designed to calculate the time-domain weight vector through double pooling channels and complete the weights of key point sequences. Furthermore, performance of the improved TDA-AMG-GCN modules has been verified on the NTU-RGB+D dataset. Its detection accuracy at the CS and CV divisions reached 84.5% and 89.8%, respectively, with an average level higher than other commonly used detection methods at present.

Exploring the impact of trip patterns on spatially aggregated crashes using floating vehicle trajectory data and graph Convolutional Networks

In recent years, increased attention has been given to understanding the spatial pattern of crashes in urban areas. Accurately capturing the spatial relationship between crash counts and variables requires extracting hidden information from multiple data sources. In this study, we propose a machine learning model to explore the spatial impact of activity patterns on spatially aggregated crash counts. Our paper introduces a two-step framework: (a) the Latent Dirichlet Allocation (LDA) model, an unsupervised method for mining hidden activity patterns from floating vehicle trajectory data, and (b) the Graph Convolutional Network (GCN) model, which builds the spatial relationship between multi-source data. The data and hidden activity patterns were aggregated into 175 Traffic Analysis Zones (TAZs) in San Francisco using spatial partitioning. The GCN model provided higher prediction accuracy than commonly used machine learning algorithms that did not consider combined spatial relationships and those that only considered traditional vehicle counts data. Furthermore, we used attribution algorithms to obtain the respective weight scores of each factor. Our results reveal that daily vehicle kilometers traveled, road density, population density, commercial activity during weekends, and residential activity during morning peak hours on weekdays are factors associated with crashes.

Soil Quality Evaluation for Cotton Fields in Arid Region Based on Graph Convolution Network

Accurate soil quality evaluation is an important prerequisite for improving soil management systems and remediating soil pollution. However, traditional soil quality evaluation methods are cumbersome to calculate, and suffer from low efficiency and low accuracy, which often lead to large deviations in the evaluation results. This study aims to provide a new and accurate soil quality evaluation method based on graph convolution network (GCN). In this study, soil organic matter (SOM), alkaline hydrolysable nitrogen (AN), available potassium (AK), salinity, and heavy metals (iron (Fe), copper (Cu), manganese (Mn), and zinc (Zn)) were determined and evaluated using the soil quality index (SQI). Then, the graph convolution network (GCN) was first introduced in the soil quality evaluation to construct an evaluation model, and its evaluation results were compared with those of the SQI. Finally, the spatial distribution of the evaluation results of the GCN model was displayed. The results showed that soil salinity had the largest coefficient of variation (86%), followed by soil heavy metals (67%) and nutrients (30.3%). The soil salinization and heavy metal pollution were at a low level in this area, and the soil nutrients and soil quality were at a high level. The evaluation accuracy of the GCN model for soil salinity/heavy metals, soil nutrients, and soil quality were 0.91, 0.84, and 0.90, respectively. Therefore, the GCN model has a high accuracy and is feasible to be applied in the soil quality evaluation. This study provides a new, simple, and highly accurate method for soil quality evaluation.

A causal-temporal graphic convolutional network (CT-GCN) approach for TBM load prediction in tunnel excavation

This research proposes a novel deep learning approach named causal-temporal graphic convolutional network (CT-GCN) which aims to provide accurate predictions on tunnel boring machine’s (TBM) load parameters. The causal associations among the key TBM operational parameters are detected and quantified with causal effects through the Peter and Clark momentary conditional independence plus (PCMCI+) method. The discovered causality can be further integrated with a deep learning model that consists of graph convolutional network (GCN) and long short-term memory (LSTM) layers for accurate prediction of TBM’s torque and thrust. Data collected from a realistic tunnel project in Singapore is utilized to demonstrate the effectiveness of the proposed approach. The load parameters are predicted with their historical values and another 7 key operational parameters. The results indicate that (1) The proposed CT-GCN approach can achieve a low mean absolute error (MAE) and root mean squared error (RMSE) of 40.90kN∙m and 64.53kN∙m for thrust force (y1) and that of 635.76kN and 1168.59kN for cutterhead torque (y2), respectively. (2) The proposed CT-GCN method achieves 12.86 % and 24.38 % average improvements in the coefficient of determination (R2) for y1 and y2, respectively when compared with the long-short term memory (LSTM) method and that of 5.62 % and 8.60 % improvements when compared with the GCN-LSTM method. (3) Compared with correlation-based GCN models, the proposed approach exerts an average improvement of 43.10 %, 41.97 %, and 22.72 % in terms of MAE, RMSE, and R2 for torque and much better performance for thrust estimation. This study contributes to improving the understanding of TBM operation by quantifying the causality among the key operational parameters. It also contributes to developing a novel deep-learning method that estimates TBM’s load parameters with high accuracy.

Capturing spatial–temporal correlations with Attention based Graph Convolutional Network for network traffic prediction

Network traffic prediction is essential and significant to network management and network security. Existing prediction methods cannot well capture the temporal–spatial correlations hidden in the network traffic and suffer from a prediction accuracy degradation for two reasons. First, the common recurrent neural network models which are leveraged to learn the temporal relations in network traffic exhibit a poor performance in long-term prediction for its limited receptive field. Second, the existing methods for modeling the spatial relationship of network traffic focus on capturing the distance relationship between nodes in the network topology, which cannot reflect the implicit correlation between network nodes. To tackle these two problems, we propose an Attention-based Graph Convolutional Network model (AGCN) for capturing both the spatial and temporal correlations in network traffic. To catch the hidden spatial dependencies in network traffic, we combine graph attention network with graph convolutional network to mine the spatial relationships of network traffic. To efficiently learn the temporal long-term relations embedded in network traffic, we design a dilated convolution module to enable an exponentially growing receptive field for handling long sequences. Experimental results on three network traffic datasets show that AGCN has excellent performance in terms of prediction accuracy and inference time compared to current mainstream methods.

Exploring the role of edge distribution in graph convolutional networks

Graph Convolutional Networks (GCNs) have shown remarkable performance in processing graph-structured data by leveraging neighborhood information for node representation learning. While most GCN models assume strong homophily within the networks they handle, some models can also handle heterophilous graphs. However, the selection of neighbors participating in the node representation learning process can significantly impact these models’ performance. To address this, we investigate the influence of neighbor selection on GCN performance, focusing on the analysis of edge distribution through theoretical and empirical approaches. Based on our findings, we propose a novel GCN model called Graph Convolution Network with Improved Edge Distribution (GCN-IED). GCN-IED incorporates both direct edges, which rely on local neighborhood similarity, and hidden edges, obtained by aggregating information from multi-hop neighbors. We extensively evaluate GCN-IED on diverse graph benchmark datasets and observe its superior performance compared to other state-of-the-art GCN methods on heterophilous datasets. Our GCN-IED model, which considers the role of neighbors and optimizes edge distribution, provides valuable insights for enhancing graph representation learning and achieving superior performance on heterophilous graphs.

Multilevel Feature Fusion-Based GCN for Rumor Detection with Topic Relevance Mining

This paper addresses the problem of detecting internet rumors in social media. Rumors do great harm to information society, making rumor detection necessary. However, existing methods for detecting rumors generally only learn pattern features or text content features from the whole propagation process, which fall short in capturing multilevel features with topic relevance of text content from social media data. In this paper, we propose a novel graph convolution network model, named multilevel feature fusion-based graph convolution network (MFF-GCN) which can employ multiple streams of GCNs to learn different level features of rumor data, respectively. We build a heterogeneous tweet graph for each single-level feature GCN to encode the topic relation among tweets based on the text contents. Experiments on real-world Twitter data demonstrate that our proposed approach achieves much better performance than the state-of-the-art methods with higher values of precision and recall as well as their corresponding F1 score. In addition, the diversity of our experimental results shows the generalization ability of our model.

An improved heterogeneous graph convolutional network for job recommendation

Job recommendation is crucial in online recruitment platforms due to the overwhelming number of job postings. Job seekers spend considerable time and effort searching for suitable employment. With millions of job seekers browsing job postings daily, the demand for accurate and effective job recommendations is more pressing than ever. To address this challenge, we propose IHGCN, an improved semi-supervised heterogeneous graph convolutional network model for job recommendation. IHGCN aims to provide job recommendations for early job seekers based on their resumes. Firstly, we introduce a novel labeling classification standard specifically tailored to early job seeker resumes. Secondly, we construct a heterogeneous resume graph where each resume is represented as a node. Job recommendation is treated as a multi-classification problem. Thirdly, our IHGCN model learns a node representation from the graph to perform effective job recommendations. To evaluate our model, we conduct experiments using a real-world resume dataset obtained from LinkedIn. The results demonstrate that IHGCN outperforms the baselines by around 10%. This study highlights the benefits of leveraging meta-paths within the Graph Convolutional Network model to address the sparsity problem caused by the one-hot representation of nodes.

Short-term passenger flow prediction based on deep learning

With the continuous improvement of urbanization, the problem of urban congestion has become increasingly prominent. Rail transit can greatly alleviate the congestion of ground traffic and is considered as a feasible solution to alleviate urban congestion. However, there are many factors affecting rail transit, among which the number of passengers accounts for the majority. Excessive passenger flow will sharply increase the probability of public safety events in confined space. The continuous development of deep learning has benefited from studies such as convolutional neural networks (CNN), so as to correctly predict short-term traffic. Despite continual model accuracy improvement, prior research has given little regard to the application boundaries of diverse networks in traffic prediction. In this paper, by comparing the results of previous studies, the prediction effect and deficiency of long short-term memory (LSTM), CNN, graph convolutional network (GCN) are discussed. And results reveal that prediction models relying on recurrent neural networks (RNN), such as LSTM, can capture the time dimension of short-term passengers very well, but not the spatial patterns, and model training time is lengthy. The GCN model would account for the spatial dependence of the traffic network, but it has poor performance when constructing the deep network. In general, each of these three models has its own unique aspects. Due to some problems in a single model, the multi-model combination deep learning framework is gradually emerging. In practical application, the complexity of the model, performance effect and application value should be considered comprehensively.

Long-Term Traffic Prediction Based on Stacked GCN Model

With the recent surge in road traffic within major cities, the need for both short and long-term traffic flow forecasting has become paramount for city authorities. Previous research efforts have predominantly focused on short-term traffic flow estimations for specific road segments and paths. However, applications of paramount importance, such as traffic management and schedule routing planning, demand a deep understanding of long-term traffic flow predictions. However, due to the intricate interplay of underlying factors, there exists a scarcity of studies dedicated to long-term traffic prediction. Previous research has also highlighted the challenge of lower accuracy in long-term predictions owing to error propagation within the model. This model effectively combines Graph Convolutional Network (GCN) capacity to extract spatial characteristics from the road network with the stacked GCN aptitude for capturing temporal context. Our developed model is subsequently employed for traffic flow forecasting within urban road networks. We rigorously compare our method against baseline techniques using two real-world datasets. Our approach significantly reduces prediction errors by 40% to 60% compared to other methods. The experimental results underscore our model's ability to uncover spatiotemporal dependencies within traffic data and its superior predictive performance over baseline models using real-world traffic datasets.

Shuffle Graph Convolutional Network for Skeleton-Based Action Recognition

A shuffle graph convolutional network (Shuffle-GCN) is proposed to recognize human action by analyzing skeleton data. It uses channel split and channel shuffle operations to process multi-feature channels of skeleton data, which reduces the computational cost of graph convolution operation. Compared with the classical two-stream adaptive graph convolutional network model, the proposed method achieves a higher precision with 1/3 of the floating-point operations (FLOPs). Even more, a channel-level topology modeling method is designed to extract more motion information of human skeleton by learning the graph topology from different channels dynamically. The performance of Shuffle-GCN is tested under 56,880 action clips from the NTU RGB+D dataset with the accuracy 96.0% and the computational complexity 12.8 GFLOPs. The proposed method offers feasible solutions for developing practical applications of action recognition.

Vismodegib Identified as a Novel COX-2 Inhibitor via Deep-Learning-Based Drug Repositioning and Molecular Docking Analysis.

Artificial intelligence algorithms have been increasingly applied in drug development due to their efficiency and effectiveness. Deep-learning-based drug repurposing can contribute to the identification of novel therapeutic applications for drugs with other indications. The current study used a trained deep-learning model to screen an FDA-approved drug library for novel COX-2 inhibitors. Reference COX-2 data sets, composed of active and decoy compounds, were obtained from the DUD-E database. To extract molecular features, compounds were subjected to RDKit, a cheminformatic toolkit. GraphConvMol, a graph convolutional network model from DeepChem, was applied to obtain a predictive model from the DUD-E data sets. Then, the COX-2 inhibitory potential of the FDA-approved drugs was predicted using the trained deep-learning model. Vismodegib, an anticancer agent that inhibits the hedgehog signaling pathway by binding to smoothened, was predicted to inhibit COX-2. Noticeably, some compounds that exhibit high potential from the prediction were known to be COX-2 inhibitors, indicating the prediction model's liability. To confirm the COX-2 inhibition activity of vismodegib, molecular docking was carried out with the reference compounds of the COX-2 inhibitor, celecoxib, and ibuprofen. Furthermore, the experimental examination of COX-2 inhibition was also carried out using a cell culture study. Results showed that vismodegib exhibited a highly comparable COX-2 inhibitory activity compared to celecoxib and ibuprofen. In conclusion, the deep-learning model can efficiently improve the virtual screening of drugs, and vismodegib can be used as a novel COX-2 inhibitor.

A systematic method of remaining useful life estimation based on physics-informed graph neural networks with multisensor data

Data-driven models, especially deep learning models, are proposed for remaining useful life (RUL) estimation with multisensor signals. Various treatments to reduce data sensitivity, addressing the difficulty of learning dynamic topologies, and coping with the lack of engineering physics guidance for model training limit the performance of these models and their use. This study proposes a systematic method to estimate RUL with multisensory data under dynamic operating conditions and multiple failure modes. Firstly, ARMA regression is introduced into the graph convolutional network(GCN) model. This allows the information loss in the GCN model following training to be lifted with low computational complexity. Secondly, the physics equations of balancing for economy and security in preventive maintenance policies is introduced in the loss function for training. This involves in a way to impose a higher penalty on delayed predictions, so to focus the neural network training on the control of high-risk situations. Finally, the method is validated on the popular C-MAPSS dataset. Compared with other cutting-edge methods, the proposed method can ensure high-fitting accuracy with strong security. In practice, the controllability and flexibility of deep learning models are enhanced, ensuring the reduction of high-risk, uncertain situations while sacrificing as little accuracy as possible.