Search

Copyright of Journal of Computer Engineering & Applications 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.)

Network representation learning (NRL) is widely used for such tasks as link prediction, node classification in network analysis. For NRL, it is a challenge in effectively fusing structural features and attribute information. To address the problem, this paper proposes a multi-order attribute network representation learning model via constructing hierarchical graphs (Multi-NRL). Firstly, the model constructs a series of hierarchical graphs on the original network through structure merging and attribute merging, which contain multi-order structural feature and attribute information from detailed to sketchy. Then, it performs hierarchical network representation on these graphs. Finally, it gains the final network representation through concatenating of the hierarchical network representation. Experimental results show Multi-NRL outperforms the best baseline by up-to 9.6% improvement in link prediction, and 13.9% in node classification with six real-world networks, which demonstrates the effectiveness of our model. [ABSTRACT FROM AUTHOR]

As a structural indicator of dense subgraphs, k-core has been widely used in community search due to its concise and efficient calculation. Many community search algorithms have been expanded on the basis of k-core. However, relevant algorithms often set k values based on empirical analysis of datasets or require users to input manually. Once users are not familiar with the graph network structure, they may miss the optimal solution due to an improper k setting. Especially in attribute social networks, characterizing communities with only k-cores may lead to a lack of semantic interpretability of communities. Consequently, this article proposes a method for identifying the optimal k-core with the greatest attribute score in the attribute social network as the target community. The difficulty of the problem is that the query needs to integrate both structural and textual indicators of the community while fully considering the diversity of attribute scoring functions. To effectively reduce computational costs, we incorporate the topological characteristics of the k-core and the attribute characteristics of entities to construct a hierarchical forest. It is worth noting that we name tree nodes in a way similar to pre-order traversal and can maintain the order of all tree nodes during the forest creation process. In such an attribute forest, it is possible to quickly locate the initial solution containing all query vertices and reuse intermediate results during the process of expanding queries. We conducted effectiveness and performance experiments on multiple real datasets. As the results show, attribute scoring functions are not monotonic, and the algorithm proposed in this paper can avoid scores falling into local optima. With the help of the attribute k-core forest, the actual query time of the Advanced algorithm has improved by two orders of magnitude compared to the BaseLine algorithm. In addition, the average F1 score of our target community has increased by 2.04 times and 26.57% compared to ACQ and SFEG, respectively. [ABSTRACT FROM AUTHOR]

Attributed network embedding aims to extract latent features of complex networks from structural topology and node attributes. Existing embedding models either use two separate learning processes to capture the complementarity of network topology and node attributes, or one process to capture their consistency. However, it is rare to simultaneously consider the complementarity and consistency of the network topology and node attributes in a single model, and network noise is often ignored. To address these issues, an unsupervised adversarial enhanced attributed network embedding model is proposed in this paper, called AEANE. The model firstly designs a novel feature extractor to simultaneously learn the consistency and complementarity features of node attributes and network topology, aiming to capture more accurate and comprehensive features. Secondly, a noisy version of the feature extractor is constructed as a fake feature generator to quickly generate more unseen fake features, aiming to simulate network noise and construct two game players. Finally, a novel adversarial auto-encoder network is designed, which integrates the auto-encoder network and the generative adversarial network into a single network, aiming to quickly learn robust node embedding and tolerate network noise. Experiments on four publicly available datasets and three analysis tasks demonstrate the advantages of our AEANE model. [ABSTRACT FROM AUTHOR]

As a structural indicator of dense subgraphs, k-core has been widely used in community search due to its concise and efficient calculation. Many community search algorithms have been expanded on the basis of k-core. However, relevant algorithms often set k values based on empirical analysis of datasets or require users to input manually. Once users are not familiar with the graph network structure, they may miss the optimal solution due to an improper k setting. Especially in attribute social networks, characterizing communities with only k-cores may lead to a lack of semantic interpretability of communities. Consequently, this article proposes a method for identifying the optimal k-core with the greatest attribute score in the attribute social network as the target community. The difficulty of the problem is that the query needs to integrate both structural and textual indicators of the community while fully considering the diversity of attribute scoring functions. To effectively reduce computational costs, we incorporate the topological characteristics of the k-core and the attribute characteristics of entities to construct a hierarchical forest. It is worth noting that we name tree nodes in a way similar to pre-order traversal and can maintain the order of all tree nodes during the forest creation process. In such an attribute forest, it is possible to quickly locate the initial solution containing all query vertices and reuse intermediate results during the process of expanding queries. We conducted effectiveness and performance experiments on multiple real datasets. As the results show, attribute scoring functions are not monotonic, and the algorithm proposed in this paper can avoid scores falling into local optima. With the help of the attribute k-core forest, the actual query time of the Advanced algorithm has improved by two orders of magnitude compared to the BaseLine algorithm. In addition, the average F1 score of our target community has increased by 2.04 times and 26.57% compared to ACQ and SFEG, respectively.

The goal of network representation learning(NRL) is embedding network nodes into low-dimensional vector space,for effective feature representation of the downstream tasks.Due to the difficulty of information collection in the real-world scene-ries,large-scale networks often meet missing links between nodes.However,the most existing NRL models are designed on the foundation of complete information networks and that causes the poor robustness in incomplete networks.To solve this problem,a deep network representation learning(DNRL) method based on incomplete information networks is proposed.Firstly,a transfer probability matrix is used to dynamically mix the structural information and attribute information to cover the excessive loss caused by incomplete structural information.Then,a deep generative model variational autoencoder with powerful feature extraction capability is used to learn low-dimensional representation of nodes,and capture the potential high nonlinear features of nodes.Compared with the commonly used network representation learning methods,the experimental results on three real attri-bute networks show that the proposed model obviously improve effect in the node classification task with different degrees of link missing,visualization results clearly demonstrate the cluster relationship of nodes.

It has become a tendency to use a combination of autoencoders and graph neural networks for attribute graph clustering to solve the community detection problem. However, the existing methods do not consider the influence differences between node neighborhood information and high-order neighborhood information, and the fusion of structural and attribute features is insufficient. In order to make better use of structural information and attribute information, we propose a model named community detection fusing graph attention network (CDFG). Specifically, we firstly use an autoencoder to learn attribute features. Then the graph attention network not only calculates the influence weight of the neighborhood node on the target node but also adds the high-order neighborhood information to learn the structural features. After that, the two features are initially fused by the balance parameter. The feature fusion module extracts the hidden layer representation of the graph attention layer to calculate the self-correlation matrix, which is multiplied by the node representation obtained by the preliminary fusion to achieve secondary fusion. Finally, the self-supervision mechanism makes it face the community detection task. Experiments are conducted on six real datasets. Using four evaluation metrics, the CDFG model performs better on most datasets, especially for the networks with longer average paths and diameters and smaller clustering coefficients. [ABSTRACT FROM AUTHOR]

Knowledge graph embedding is aimed at capturing the semantic information of entities by modeling the structural information between entities. For long-tail entities which lack sufficient structural information, general knowledge graph embedding models often show relatively low performance in link prediction. In order to solve such problems, this paper proposes a general knowledge graph embedding framework to learn the structural information as well as the attribute information of the entities simultaneously. Under this framework, a H-AKRL (Hypergraph Neural Networks based Attribute-embodied Knowledge Representation Learning) model is put forward, where the hypergraph neural network is used to model the correlation between entities and attributes at a higher level. The complementary relationship between attribute information and structural information is taken full advantage of, enabling H-AKRL to finally achieve the goal of improving link prediction performance. Experiments on multiple real-world data sets show that the H-AKRL model has significantly improved the link prediction performance, especially in the embeddings of long tail entities. [ABSTRACT FROM AUTHOR]

Visual Question Answering (VQA) is a research topic of significant interest at the intersection of computer vision and natural language understanding. Recent research indicates that attributes and knowledge can effectively improve performance for both image captioning and VQA. In this article, an inner knowledge-based Img2Doc algorithm for VQA is presented. The inner knowledge is characterized as the inner attribute relationship in visual images. In addition to using an attribute network for inner knowledge-based image representation, VQA scheme is associated with a question-guided Doc2Vec method for question--answering. The attribute network generates inner knowledge-based features for visual images, while a novel question-guided Doc2Vec method aims at converting natural language text to vector features. After the vector features are extracted, they are combined with visual image features into a classifier to provide an answer. Based on our model, the VQA problem is resolved by textual question answering. The experimental results demonstrate that the proposed method achieves superior performance on multiple benchmark datasets. [ABSTRACT FROM AUTHOR]

Existing network embedding models are mostly designed for homogeneous networks or heterogeneous networks, but ignore the special features of bipartite network which arise in recommender systems, search engines, question answering systems and so on. Meanwhile bipartite networks mostly include rich attribute information. To address the above challenging issues, this paper proposes ABNE (attributed bipartite network embedding). Specifically, ABNE first preserves the explicit relations in the bipartite network by decomposing edges into sets of indirect relationships between neighborhood nodes. Then it calculates attribute similarity matrix by cosine similarity and as part of the weight matrix to guide the biased random walk to embed implicit relations and attribute information. Finally, ABNE introduces an optimization framework to obtain the node representation vector which carries both structure information and attribute information. Several tasks have been conducted on four public datasets and compared with other state-of-the-art embedding models. The experimental results show superiority and rationality of ABNE model.

In attribute network, the attribute information associated with nodes is essential to improve the performance of network embedding tasks. Nevertheless, network is a graph structure, in which nodes not only contain attribute information but also embrace the rich structural information. In order to make full use of the structural information, firstly, this paper defined influential node characteristics, spatial relationships, and constructed similarity matrix based on the definition of link prediction. Then it mapped correlation similarity vector associated with nodes in the binary group to the relationship space of the adjacency matrix, so as to maintain the node vector matrix structure information feature. Based on the definition of normalized graph Laplacian, it fused the attribute information and label feature and integrated the above three kinds of information into an optimization framework. Finally, it inferenced the projection matrix by calculating the local maximum value through a second order derivative. Experimental results indicate that the proposed algorithm can effectively utilize information of the adjacency structure with the binary group of nodes, and compared with other benchmark network embedding algorithms, it also can achieve better results on the node classification task. [ABSTRACT FROM AUTHOR]

Due to the lack of deep-level information mining and utilization in the existing network representation learning methods,the potential pattern structure similarity was proposed by further exploring the potential information in the network.The similarity score between network structures was defined to measure the similarity between various structures so that nodes could cross irrelevant vertices to obtain high-order similarities on the global structure.In order to achieve the best effect,deep learning was used to fuse multiple information sources to participate in training together to make up for the deficiency of random walks.In the experiment,Lap,DeepWalk,TADW,SDNE and CANE were selected as comparison methods,and three real-world networks were used as data sets to verify the validity of the model,and experiments of node classification and link reconstruction are carried out.In the node classification,the average performance is improved by 1.7 percentage points for different datasets and training proportions.In the link reconstruction experiment,only half the dimension is needed to achieve better performance.Finally,the performance improvement of the model under different network depths was discussed.By increasing the depth of the model,the average performance of node classification increased by 1.1 percentage points.

Attribute network anomaly detection has attracted more and more research attention due to its wide application in social media, financial transactions, and network security. However, most of the existing methods only consider the network structure or attribute information to detect anomalies, ignoring the combined information of the node structure and attributes in the network. A novel anomaly detection method in attributed networks based on walking autoencoder named RW2AEAD is proposed in this paper, considering structure and attribute information. Besides capturing the network's structural information by random walking, it gets the combined information of structures and the attributes that are closely related to the structures. And then, the structure and combined reconstruction error of node are obtained by inputting into the autoencoder composed of SkipGram and CBOW. In addition, the global attribute reconstruction error of the node is obtained through the multi-layer attribute autoencoder. Finally, the anomaly score of the node comprehensively considers the above three reconstruction errors, and detects anomalous nodes by setting the threshold and the score ranking. Experiments show that the performance of the proposed RW2AEAD is better than other baseline algorithms in four real datasets. [ABSTRACT FROM AUTHOR]

It has become a tendency to use a combination of autoencoders and graph neural networks for attribute graph clustering to solve the community detection problem. However, the existing methods do not consider the influence differences between node neighborhood information and high-order neighborhood information, and the fusion of structural and attribute features is insufficient. In order to make better use of structural information and attribute information, we propose a model named community detection fusing graph attention network (CDFG). Specifically, we firstly use an autoencoder to learn attribute features. Then the graph attention network not only calculates the influence weight of the neighborhood node on the target node but also adds the high-order neighborhood information to learn the structural features. After that, the two features are initially fused by the balance parameter. The feature fusion module extracts the hidden layer representation of the graph attention layer to calculate the self-correlation matrix, which is multiplied by the node representation obtained by the preliminary fusion to achieve secondary fusion. Finally, the self-supervision mechanism makes it face the community detection task. Experiments are conducted on six real datasets. Using four evaluation metrics, the CDFG model performs better on most datasets, especially for the networks with longer average paths and diameters and smaller clustering coefficients.

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.)

Copyright of Journal of Computer-Aided Design & Computer Graphics / Jisuanji Fuzhu Sheji Yu Tuxingxue Xuebao is the property of Gai Kan Bian Wei Hui 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.)

Due to the lack of deep-level information mining and utilization in the existing network representation learning methods, the potential pattern structure similarity was proposed by further exploring the potential information in the network. The similarity score between network structures was defined to measure the similarity between various structures so that nodes could cross irrelevant vertices to obtain high-order similarities on the global structure. In order to achieve the best effect, deep learning was used to fuse multiple information sources to participate in training together to make up for the deficiency of random walks. In the experiment, Lap, DeepWalk, TADW, SDNE and CANE were selected as comparison methods, and three real-world networks were used as data sets to verify the validity of the model, and experiments of node classification and link reconstruction are carried out. In the node classification, the average performance is improved by 1.7 percentage points for different datasets and training proportions. In the link reconstruction experiment, only half the dimension is needed to achieve better performance. Finally, the performance improvement of the model under different network depths was discussed. By increasing the depth of the model, the average performance of node classification increased by 1.1 percentage points. [ABSTRACT FROM AUTHOR]

The network embedding method via deep auto-encoder can capture and preserve the structure of highly non-linear network. But when the links are sparse, the representation is not accurate. In order to improve the accuracy, this paper proposed LEANE model. More specifically, this paper firstly used multiple deep auto-encoders to learn a representation of the lowdimensional network where preserved the network topology and semantic information on nodes of networks. Then it introduced a pairwise constraint on the low-dimensional representation according to the similarity between two nodes to enhance local network structures. The pairwise constraint considered both topological neighbor relationship and semantic similarity of nodes. So, the LEANE model can better preserve the original network topology and semantic in-formation. Extensive experiments on five realworld networks demonstrate a superior performance of this method over state-of-the-art methods for embedding. [ABSTRACT FROM AUTHOR]

Network embedding aims to learn low-dimensional representations for nodes in social networks, which can serve many applications, such as node classification, link prediction and visualization. Most of network embedding methods focus on learning the representations solely from the topological structure. Recently, attributed network embedding, which utilizes both the topological structure and node content to jointly learn latent representations, becomes a hot topic. However, previous studies obtain the joint representations by directly concatenating the one from each aspect, which may lose the correlations between the topological structure and node content. In this paper, we propose a new attributed network embedding method, TLVANE, which can address the drawback by exploiting the deep variational autoencoders (VAEs). Particularly, a two-level VAE model is built, where the first-level accounts for the joint representations while the second for the embeddings of each aspect. Extensive experiments on three real-world datasets have been conducted, and the results demonstrate the superiority of the proposed method against state-of-the-art competitors. [ABSTRACT FROM AUTHOR]

In recent years, evolutionary multi-objective based community detection algorithms are widely used in attribute networks, but such algorithms usually ignore the attributes between nodes and may lead to incorrect node division. Therefore, this paper proposes an evolutionary multi-objective attribute community detection based on similarity fusion strategy with central nodes. First, this paper proposes a pre-processing of similarity fusion to completely utilize node and topology information, the topological similarity matrix of the network is effectively combined with the attribute similarity matrix to obtain the fusion similarity matrix, and finds central nodes based on node assignment of the fusion similarity matrix. Then, the pre-division set of the attribute network is selected by central nodes and the label update equation is designed. In the scheme of evolutionary algorithm, using the community results initialized for the network after the label update can speed up the iterative process of the algorithm. Finally, a modularity-based community integration strategy is proposed to correct community detection results of attribute network based on modularity of neighbor nodes. Comparing the effectiveness of the proposed algorithm with four excellent community detection algorithms for attribute networks on fifteen real networks and six synthetic networks, the proposed algorithm can achieve high division accuracy in most networks. • A similarity fusion strategy designed by combining the topological and attribute similarity to find central nodes. • The pre-division set of attribute network is selected based on the central nodes and the label update equation is designed. • A modularity-based correction method is proposed based on the neighbors of nodes and community integration strategy. [ABSTRACT FROM AUTHOR]

Abstract Community detection is an important problem for understanding the structure and function of complex networks and has attracted a lot of attention in recent decades. Most community detection algorithms only focus on the topology of networks. However, there is still much valuable information hidden in the networks, such as the attributes or content of the nodes and the useful prior information. Obviously, taking full advantage of these resources can improve the effectiveness of community detection. In this paper, we present a semi-supervised community detection framework named SCDAN (Semi-supervised Community Detection in Attribute Networks), in which a non-negative matrix factorization model is utilized to effectively integrate network topology, node attributes and individual labels simultaneously. The comparative experiments on real-world networks show that SCDAN significantly improves the performance of community detection and provides semantic interpretation of communities. Highlights • A new framework integrating the topology, attributes and individual labels effectively. • Utilize a non-negative matrix factorization model to establish the framework. • Derive the multiplication update rules to optimize the objective function. • Extensive experiments show the effectiveness of our method. [ABSTRACT FROM AUTHOR]

In recent years, complex networks have become increasingly difficult to detect effectively with the rapid development of networks. On the one hand, the scale of the network has increased sharply, and on the other hand, the nodes in the network contain rich content. The existing algorithms do not take the different richness of attribute information of nodes in the attribute network into account. And the topology information of the node is inconsistent with the attribute information. Thus, this paper proposes an attribute community detection algorithm based on latent representation learning and graph-regularized non-negative matrix factorization (LRL-GNMF). First, the topological information and attribute information in the attribute network is decomposed based on non-negative matrix factorization respectively. Thereby, the member distribution matrix and the attribute distribution matrix are obtained. Secondly, an affinity matrix is constructed for the attribute matrix, and the latent representation of the attribute information is obtained using the latent representation learning method. In addition, a transition matrix is constructed according to the Markov transition probability. The node membership distribution matrix and the attribute distribution matrix are linked. Finally, since the attribute information of nodes in the network is different, a topology-dominated model and an attribute-dominated model are respectively constructed to solve this problem. At the same time, a graph regularization term is introduced to guide the model to obtain more accurate community detection. This paper conducts experimental analysis on 8 real networks. The experimental results show that the proposed algorithm in this paper outperforms the other 11 compared algorithms in the evaluation indicators of community detection accuracy and standard mutual information. • An affinity matrix is constructed for the attribute information to represent the node association in the attribute information. • The NMF method is used to model the affinity matrix and the potential representation of the attribute matrix is obtained. • A transition matrix is designed to establish a mapping relationship, which can decompose the community results. • The topology information and attribute information are modeled separately. • The transfer matrix is used to link the two kinds of information, and the models dominated by topology and attribute information are established respectively. [ABSTRACT FROM AUTHOR]