Search

Entity relation extraction aims to identify the semantic relation between entity pairs in a sentence and is an important technical support for downstream tasks such as question answering systems and semantic searching. Existing relation extraction models mainly rely on neural networks to extract the semantic information of sentences, ignoring the critical role of important phrase information in relation extraction. For this problem, this paper proposes a relation extraction model based on BERT gated multi-window attention network (BERT-GMAN). The model first uses BERT to extract the semantic representation features of the sentence and its constraint information. Secondly, it constructs the key phrases extraction network to obtain multi-granularity phrase information and uses element-wise max pooling to select key phrases features. Thirdly, it adopts classification feature perception network to further filter and globally perceive key phrase feature to form the overall features of relation classification. Finally, it combines with Softmax classifier to perform relation extraction. The experimental results on the Semeval-2010 Task 8 dataset show that the performance of the model in this paper is further improved compared with the existing methods, and the F1-score reaches 90.25%. [ABSTRACT FROM AUTHOR]

Dynamic networks are popularly used to describe networks that change with time. Although there have been a large number of research works on understanding dynamic networks using link prediction, node classification and community detection, there is rare work that is specially designed to address the challenge of big network size of dynamic networks. To this end, we study in this paper an emerging and challenging problem of network coarsening in dynamic networks. Network coarsening refers to a class of network "zoom-out" operations where node pairs and edges are grouped together for efficient analysis on big networks. However, existing network coarsening approaches can only handle static networks where network structure weights have been predefined before the coarsening calculation. Under the observation that big networks are highly dynamic and naturally change over time, we consider in this paper to embed information diffusion data which reflect the dynamics of networks for network coarsening. Specifically, we present a new Semi-NetCoarsen approach that jointly maximizes the likelihood of observing the information diffusion data and minimizes the network regularization with respect to the predefined network structural data. The learning function is convex and we use the accelerated proximal gradient algorithm to obtain the global optimal solution. We conduct experiments on two synthetic and five real-world data sets to validate the performance of the proposed method. [ABSTRACT FROM AUTHOR]

Multi-view clustering has gained soar attention in the field of data mining in recent years. Most multi-view clustering methods are considered to be complete for each view. However, in many applications, some views may contain missing instances, resulting in the incomplete multi-view learning problem. Incomplete views cannot be directly dealt with traditional multi-view clustering methods. In this paper, we propose a One-step multi-view subspace clustering with incomplete views (OMVSC-IV) method based on low-rank matrix factorization to overcome this problem. Specifically, the proposed algorithm uses low-rank matrix factorization to learn a consensus representation matrix, and then combined with the objective function of non-negative embedding and spectral embedding subspace clustering proposed in this paper. The whole process is established into the same objective function for joint optimization and requires no post-processing (e.g., K -means), thus avoiding the defect of being sensitive to initial values. Furthermore, the accuracy, parameter, sensitivity and convergence of the algorithm are studied by experiments. Experimental results on nine data sets show that the proposed algorithm performs superior to some of the latest algorithms. [ABSTRACT FROM AUTHOR]

In most Internet of Things (IoT) scenes, data missing can be unavoidable when huge number of smart devices are collecting data uninterruptedly. Therefore, data imputation can be an integral part of pre-processing before data mining. It is widely known that IoT time series show strong dependencies in both spatial and temporal dimension, and the spatial relation among the devices is in non-euclidean space. However, most machine-learning-based and deep-learning-based approaches either only take temporal features into account or only catch spatial features in euclidean space. In this paper, we propose a novel network as ST-VAE (Spatio-Temporal Variational Auto-Encoder) to address the problem above. Our architecture is mainly based on Variational Auto-Encoder (VAE). Specifically, two kinds of VAE are utilized. One is for calculating the adjacent matrix of device network which is the essential input of GCN, and the other is for data imputation task based on the spatial and temporal dependencies. Experiments conducted on different real-world and public datasets demonstrate that our ST-VAE can not only populate the missing spatio-temporal data accurately but also outperforms other state-of-art approaches from the whole. [ABSTRACT FROM AUTHOR]

In the past few decades, numerous studies have demonstrated the effectiveness of diet modification in preventing and controlling different chronic diseases, bringing increased attention to the creation of nutritional recipes. Currently, the development of recipes is primarily based on personal experience rather than nutritional specifications. To this end, this paper first uses kernel canonical correlation analysis to demonstrate the relationship between nutrients in food and recipes. Based on this relationship, a new recipe expression method is put forward, which objectively reflects differing importance of an ingredient in differing recipes. Then, recipes are composed using an auto-encoder in deep learning, and a fusion model of two auto-encoders is proposed for better concocting recipes. This paper uses two machine learning methods, namely, non-negative matrix factorization and two-step regularized least squares, to form recipes. To tackle overfitting and instability in non-negative matrix factorization during the training of recipe model, we introduce the Frobenius norm to redefine the objective function and add non-smooth sparse matrices. Similar food has similar taste, but their nutrients might differ. This paper also considers nutrients as a kernel matrix of the two-step regularized least squares, which can effectively avoid the occurrence of different food combinations with similar taste. Experimental results show that developing recipes based on nutrients in food is feasible and effective in the context of machine learning. [ABSTRACT FROM AUTHOR]

Signed network embedding is an approach to learning low-dimensional representations of nodes in signed networks with both positive and negative links, which facilitates downstream tasks such as link prediction with general data mining frameworks. Due to the distinct properties and significant added value of negative links, existing signed network embedding methods usually design dedicated methods based on social theories such as balance theory and status theory. However, existing signed network embedding methods ignore the characteristics of multiple facets of each node and mix them up in one single representation, which limits the ability to capture the fine-grained attentions between node pairs. In this paper, we propose MUSE , a MU lti-faceted attention-based S igned network E mbedding framework to tackle this problem. Specifically, a joint intra- and inter-facet attention mechanism is introduced to aggregate fine-grained information from neighbor nodes. Moreover, balance theory is also utilized to guide information aggregation from multi-order balanced and unbalanced neighbors. Experimental results on four real-world signed network datasets demonstrate the effectiveness of our proposed framework. [ABSTRACT FROM AUTHOR]

• Both labeled data and unlabeled images are used to reduce the cost of labeling the flame hole image artificially. Making full use of the efficiency of CNN feature extraction and the comprehensiveness of fusion, as well as the high computation efficiency of ELM training process. • Laplacian regularization is utilized in the paper for local geometry information extraction, the proposed CNN-LapsELM can fully excavate information which contained in all available data. • The fusion of original calculated features for top-level ELM training process can greatly improve the accuracy for SD identification, and the experiment result indicates the SD accuracy is high to 87%. The superheat degree (SD) in industrial aluminum electrolysis cell is a critical index that can maintain the energy balance, improve the current efficiency and improve production. However, the existing SD identification is mainly relying on artificial experience and the accuracy of SD is far from satisfactory. Further, artificial costs and physical equipment are expensive and time-consuming. In this paper, we propose a deep soft sensor method for SD detection. First, CNN is utilized for flame hole image feature extraction. Second, a semi-supervised extreme learning machine (ELM) that integrates Laplacian regularization is further used for SD classification. The main contributions of the paper are: (1) The proposed CNN-LapsELM utilizes the CNN for flame hole image feature extraction and then ELM for further classification, which fully takes advantage of CNN's ability for complex feature extraction, ELM's excellent generalization ability, and high computation efficiency. (2) Both the labeled and unlabeled samples are utilized for the CNN-LapsELM training process. It fully leverages the information contained in unlabeled data. At the same time, Laplacian regularization is utilized for learning the manifold structure of hole image samples, so the performance of the proposed CNN-LapsELM are improved. (3) The proposed CNN-LapsELM algorithm improves the generalization ability and robustness. The comparison result demonstrates that the CNN-LapsELM is superior to the existing SD identification and the accuracy is 87%. [ABSTRACT FROM AUTHOR]

• A novel anomalous subgraph detection method fuses the multiple experts' interactions. • Maximizing the non-parametric scan statistic over the connected subgraphs. • Detecting the subgraph as an event by the data-driven and user-driven procedures. • Guaranteeing to get the globally optimal solution without disconnecting the subgraph. • Achieving better performance than representative baselines on three datasets. Event detection on networks is an important research task in data mining. Most previous approaches usually detect a particular type of event that satisfies the predefined rules in the model, i.e., data-driven methods. However, few methods consider the diverse interests of multiple experts during the detection process to discover unexplored events. In this paper, we regard interactive event detection on attributed networks as Multi-users Interaction Anomalous Subgraph Detection (MIASD), where events are represented as connected subgraphs on networks. The core of our method is automatically maximizing a non-parametric scan statistic over connected subgraphs to identify the most anomalous subgraph as the event and integrating the interactions from multiple experts simultaneously, i.e., under both data-driven and user-driven paradigm. In our approach, we first define individual and shared environments of experts on attributed networks to model the interests of experts, including the interactive operations and Recommended Interaction Domain. Afterwards, we propose an efficient anomalous subgraph detection algorithm in which each expert gives feedback separately on the vertices in the Recommended Interaction Domain; then, the non-parametric scan statistic over the connected subgraph is approximately maximized by integrating various feedback from experts into the iterative subgraph expansion process. In this way, our method retrieves the most anomalous subgraph on the network as the final event, which contains the potential unexplored information. We have conducted extensive experiments on three real-world datasets and proved that our algorithm could achieve better performance compared with several competitive baselines. Furthermore, the case study shows that our method could detect global abnormal events effectively. [ABSTRACT FROM AUTHOR]

Spectral Clustering (SC) is an important method in areas such as data mining, image processing, computer science and so on. It attracts more and more attention owing to its effectiveness in unsupervised learning. However, SC has poor performance in the high-dimensional data. Traditional SC methods conduct the spectral embedding of the affinity matrix among data at the first beginning, and then obtain clustering results by the K -means clustering. The also have drawbacks int two processing steps: the clustering results are sensitive to the affinity matrix which may be inaccurate and the post-processing K -means may also be limited by its initialization problem. In the paper, a new approach which joints spectral embedding and clustering with structured graph optimization (called JSEGO) is proposed. In the new model, the low-dimensional representation of data can first be obtained by the spectral embedding method, which can handle with the high-dimensional data better. Then, the optimization similarity matrix would be obtained with such the embedded data. Furthermore, the learning structure graph gives feedback to the similarity matrix to generate better spectral embedded data. As a result, better similarity matrix and clustering result can be obtained by the iterations simultaneously, which are often conducted in two separate steps in the spectral clustering. As a result, the drawbacks introduced by the two processing introduces can be solved. At last, we use an alternative optimization method in the new model and conduct the theoretical analysis by comparing this proposed method with K -means clustering. Experiments based on synthetic data and actual benchmark data prove the advantage of this new approach. [ABSTRACT FROM AUTHOR]

Due to the lack of adequate training data and sufficient mining of prior knowledge related to perceived quality, most existing image quality assessment (IQA) methods show limited generalization performance. In this paper, we study the prior knowledge from the factors affecting perceived quality, and introduce a novel progressive multi-task learning based blind IQA method. Inspired by the definition of IQA: human comprehensive perception for degradation of image content, we firstly deconstruct IQA into three elements, i.e., image content, pattern of degradation, and intensity of degradation. Based on these elements, we design the corresponding auxiliary tasks for instructing the network to learn IQA. By statistical analysis on a great deal of data, we find that there is progressive relevance among the four tasks. Furthermore, we mathematically derive that introducing the progressive relevance into a multi-task learning network can effectively constrain the hypothesis space of the main task. Under the guidance of the derivation, we propose an end-to-end IQA framework based on progressive multi-task learning. Experimental results demonstrate the excellent generalization capability of the proposed method, which achieves state-of-the-art performance against these existing IQA methods. [ABSTRACT FROM AUTHOR]

Anomalies affect data quality and lead to unexpected analysis results in data mining. Although techniques to handle anomalies do exist, they can fail in considering density changes of object along with incremental neighbors. This paper proposes an outlier factor based on the change of adjoint dynamical kernel density (ADD) to represent the degree of the object being an anomaly. The factor is equal to the ratios of the adjoint dynamical kernel density fluctuation (ADDF) of the object and the average ADDF of its neighborhood. ADDF is estimated by the difference of ADD, which describes the difference of every two consecutive adjoint kernel densities (AKD) of the object. AKD indicates kernel densities of the object while its neighbors are added one by one. Importantly, the kernel function is adopted to measure the distance between objects where the kernel trick improves discriminability between objects and reduces the computational burden of the algorithm. The experiments are performed on eight datasets to evaluate the effectiveness of the proposed method with different kernel functions. The experimental results have shown that the proposed method with the Gaussian kernel function has better performance of anomalies recognition and higher adaption of the parameter k of k -nearest neighbors over some other anomaly detection methods. [ABSTRACT FROM AUTHOR]

Class imbalance learning is one of the most important topics in the field of machine learning and data mining, and the Synthetic Minority Oversampling Techniques (SMOTE) is the common method to handle this issue. The main shortcomings of the classic SMOTE and its variants is the interpolation of potential noise and unrepresentative examples. This paper is devoted to proposing a novel parameter-free SMOTE mechanism to produce sufficient representative synthetic examples while avoiding interpolating noisy examples. Specifically, two types of minority class examples are defined, namely boundary and safe minority examples. The synthetic examples generation procedure fully reflects the characteristics of the minority class examples with filling the region dominated by the minority class and expanding the margin of the minority class. To verify the effectiveness and robustness of the proposed method, a thorough experimental study on forty datasets selected from real-world applications is carried out. The experimental results indicate that our proposed method is competitive to the classic SMOTE and its state-of-the-art variants. [ABSTRACT FROM AUTHOR]

Imbalanced classification is a challenging task in the fields of machine learning and data mining. Cost-sensitive learning can tackle this issue by considering different misclassification costs of classes. Weighted extreme learning machine (W-ELM) takes a cost-sensitive strategy to alleviate the learning bias towards the majority class to achieve better classification performance. However, W-ELM may not achieve the optimal weights for the samples from different classes due to the adoption of empirical costs. In order to solve this issue, multi-objective optimization-based adaptive class-specific cost extreme learning machine (MOAC-ELM) is presented in this paper. To be specific, the initial weights are first assigned depending on the class information. Based on that, the representation of the minority class could be enhanced by adding penalty factors. In addition, a multi-objective optimization with respect to penalty factors is formulated to automatically determine the class-specific costs, in which multiple performance criteria are constructed by comprehensively considering the misclassification rate and generalization gap. Finally, ensemble strategy is implemented to make decisions after optimization. Accordingly, the proposed MOAC-ELM is an adaptive method with good robustness and generalization performance for imbalanced classification problems. Comprehensive experiments have been performed on several benchmark datasets and a real-world application dataset. The statistical results demonstrate that MOAC-ELM can achieve competitive results on classification performance. [ABSTRACT FROM AUTHOR]

• Subset-based feature selection + multi-layer perceptron. • Subset-based feature selection + radial basis function neural networks. • Semi-wrapper feature subset selection approach based on Naïve Bayes. • Extensive experiments on 34 data sets (22 from University of California at Irvine –UCI- and 12 synthetic data sets). Data preparation via feature selection + neural networks. • Feature selection: 7 approaches, which are 5 filters and 2 semi-wrappers (based on Correlation, consistency and Naïve Bayes; other with SOAP or SU), have been tested on difficult classification problems. This paper explores widely the data preparation stage within the process of knowledge discovery and data mining via feature subset selection in the context of two very well-known neural models: radial basis function neural networks and multi-layer perceptron. It is known the best performance of wrapper attribute selection methods based on the evaluation measure provided by a classifier, although the temporal complexity of learning neural networks practically precludes the use of wrapper techniques, especially in complex databases with high dimensionality and a large number of labels. In this paper, we propose the use of the Naïve Bayes classifier as a fitness function within a semi-wrapper feature selection approach. The Naïve Bayes classifier is a good fast approach to a neural network and utilising it as a measure of goodness in a backward search on a ranking provides a specific attribute selection method for neural networks in complex data. The test-bed consists of 34 binary and multi-class classification problems and 7 feature selectors. Of these, there are 6 data sets with upwards of 5 classes. According to the reported accuracy results that have been supported by non-parametric statistical tests in different scenarios, our method has been shown to be very suitable for both kinds of neural networks. Moreover, the reduced feature-space is around 20% of the full attribute space. The speedup with the aforementioned semi-wrapper is very outstanding and its value fluctuates, on average, from about 1.5 with radial basis function neural networks to around 30 with multi-layer perceptron. [ABSTRACT FROM AUTHOR]

Abstract Currently, knowledge discovery in databases is an essential first step when identifying valid, novel and useful patterns for decision making. There are many real-world scenarios, such as bankruptcy prediction, option pricing or medical diagnosis, where the classification models to be learned need to fulfill restrictions of monotonicity (i.e. the target class label should not decrease when input attributes values increase). For instance, it is rational to assume that a higher debt ratio of a company should never result in a lower level of bankruptcy risk. Consequently, there is a growing interest from the data mining research community concerning monotonic predictive models. This paper aims to present an overview of the literature in the field, analyzing existing techniques and proposing a taxonomy of the algorithms based on the type of model generated. For each method, we review the quality metrics considered in the evaluation and the different data sets and monotonic problems used in the analysis. In this way, this paper serves as an overview of monotonic classification research in specialized literature and can be used as a functional guide for the field. [ABSTRACT FROM AUTHOR]

Abstract In the era of big data, all types of data with increasing samples and high-dimensional attributes are demonstrating their important roles in various fields, such as data mining, pattern recognition and machine learning, etc. Meanwhile, machine learning algorithms are being effectively applied in large-scale data processing. This paper mainly reviews the classical dimensionality reduction and sample selection methods based on machine learning algorithms for large-scale data processing. Firstly, the paper provides a brief overview to the classical sample selection and dimensionality reduction methods. Then, it pays attention to the applications of those methods and their combinations with the classical machine learning methods, such as clustering, random forest, fuzzy set, and heuristic algorithms, particularly deep leaning methods. Furthermore, the paper primarily introduces the application frameworks that combine sample selection and dimensionality reduction in the context of two aspects: sequential and simultaneous, which almost all get the ideal results in the processing of the large-scale training data contrasting to the original models. Lastly, we further conclude that sample selection and dimensionality reduction methods are essential and effective for the modern large-scale data processing. In the future work, the machine learning algorithms, especially the deep learning methods, will play a more important role in the processing of large-scale data. [ABSTRACT FROM AUTHOR]

Document-level event argument linking aims to find global event arguments to fill an event's semantic role, which is a challenging task owing to the appearance of long contexts and the issue of data sparsity. In this paper, we study a new formulation to address the above challenges in document-level EAL, by explicitly framing the task as a machine reading comprehension (MRC) problem. In this formulation, argument extraction is viewed as a question answering procedure. To better transfer each semantic role into a question, we propose a back-translation based query generation method, which can effectively generate well-formed questions without adopting huge human effort. Moreover, to better capture the non-local dependencies between triggers and arguments, we devise a dependency-guided question answering process, which can explore the underlying structure of the document to boost learning. The extensive experiments on a benchmark have justified the effectiveness of our approach. Particularity, our approach achieves substantially improvement over previous methods, leading to +5.7% in F1 in the full argument linking setting. Moreover, our approach is particular data-efficient and demonstrates superior performance in the data-low scenario with limited training data. [ABSTRACT FROM AUTHOR]

Edge computing is an emerging computing paradigm, which offers a great opportunity to implement data mining-based services and applications for a large number of devices and sensors in Internet of Things. However, the new paradigm is faced with security and privacy challenges due to the diversity and the limited capability of edge components. In particular, data privacy is one of the most concerned problems for all the participants. In this paper, we propose a framework of privacy-preserving data mining based on private random decision trees in edge computing, which not only gives the strong privacy guarantee, but also provides a certain amount of data utility. Firstly, we design a preservation framework to implement private random decision trees satisfying local differential privacy. Secondly, we present the concrete implementations of algorithms and the corresponding task that each participant needs to undertake. Thirdly, we analyze the key factors to influence privacy and utility, including the allocation of data and privacy budget. Fourthly, we give the improved algorithms to further increase the utility with strong privacy preservation. Finally, extensive experiments demonstrate the good performance of our designed framework. [ABSTRACT FROM AUTHOR]

Information extraction (IE) is a crucial task in natural language processing, focusing on classifying and structuring informative elements from textual data. Despite significant performance improvements in recent years, existing methods face challenges in terms of scalability, adaptability, and explainability when applied in real-world scenarios. This paper presents a pioneering approach, namely the KNN-IE framework, tailored to address this challenge in key IE tasks. Diverging from conventional KNN methods primarily employed in classification tasks based on point-to-point similarity, our framework introduce an instance-to-instance similarity, called structural similarity, where each instance can contain multiple points as its elements. To provide supervision signals during training and enable better prediction explanations, we define a text-based measurement between two instances which is reasonable and can be calculated before trainingIn the training phase, we leverage structure-aware contrastive learning to help model learn structural similarity between different instances, especially those with the same type. During inference, schema-constraint bipartite matching between target text and retrieved instances enables KNN-IE to concurrently tackle the classification and structuring of informative elements, in a single round of retrieval process. Every time after finishing extraction, those retrieved structurally similar instances can work as well comprehensible evidences to help user understand why KNN-IE makes such predictions. Experimental evaluations across four prevalent IE tasks and ten diverse datasets demonstrate that KNN-IE can compete with state-of-the-art IE methods in fully supervised settings. Other results further underscore its superior scalability, adaptability, and explainability, affirming its potential for effective deployment in practical applications. [Display omitted] • Introduce KNN-IE: Innovative KNN IE method for simultaneous classification and structuring. • KNN-IE excels in scalability, adaptability, and explainability for real-world IE systems. • Extensive experimental results on 10 datasets and four typical IE tasks validate KNN-IE's effectiveness. [ABSTRACT FROM AUTHOR]

Outlier detection is one of the core problems in the field of data mining. Graph-based approaches are widely acknowledged for their robust performance in outlier detection, yet constructing a comprehensive graph to represent the original data remains challenging. Some methods rely solely on the global distribution of the data, leading to low detection accuracy; in addition, when incorporating the local information of the data, the problem of dangling links may arise. To improve the graph construction process, this paper proposes an outlier detection algorithm based on a hybrid graph (HGOD). First, the neighbors of each object are obtained in an adaptive manner, and a neighborhood graph is created. Subsequently, a global connectivity graph is constructed using a minimum spanning tree. Then, these two graphs are merged into a hybrid graph, and Markov random walks are conducted on it. Finally, outliers are identified based on the converged stationary distribution. The performance of HGOD is evaluated on both real-world and synthetic datasets. The precision and AUC of HGOD algorithm surpasses that of the other seven algorithms. Ablation and analysis experiments further demonstrate that the HGOD algorithm is effective and has outstanding performance in outlier detection. [ABSTRACT FROM AUTHOR]

Matrix completion is widely used in many practical applications such as computer vision and data mining. In this paper, we consider the following two issues arising in real scenarios. First, the collected datasets are damaged by sparse noise and column outliers simultaneously; second, the datasets are not static in nature due to the existing of out-of-sample. Both of them have been ignored by most existing methods. In contrast with the traditional matrix completion algorithms which aim at recovering the entire matrix directly, we in this paper aim to first learn a low dimensional subspace by recovering a subset of collected samples, and then utilize it to estimate the missing values of residual data. There are two important advantages about this transformation. First, weakening the deviation caused by column outliers. Second, providing a direction for efficiently solving the out-of-sample problem. Particularly, to further improve the robustness of proposed method to sparse noise, we present a novel robust matrix completion model and a robust vector completion model, and both of them are based on non-convex ℓ p -norm (0 <  p  < 1). In experiments, the proposed method and other state-of-the-art algorithms will be used to cope with two problems: matrix completion and vector completion. Numerical results on real datasets and artificial datasets demonstrate that our method can provide a significant performance advantage over alternative methods. [ABSTRACT FROM AUTHOR]

Neural Networks and Support Vector Machines (SVMs) are two of the most popular and efficient supervised classification models. However, in the context of large datasets many complexity issues arise due to high memory requirements and high computational cost. In the context of the application of Data Mining algorithms, data reduction techniques attempt to reduce the size of training datasets in terms of the number of instances by selecting some of the existing instances or by generating new training instances. The idea is to speed up the application of the data mining algorithm with minimum or no sacrifice in performance. Data reduction techniques have been extensively used in the context of k -Nearest Neighbor classification, a lazy classifier that works by directly using a training dataset rather than building a model. This paper explores the application of data reduction techniques as a preprocessing step before the training step of Neural Networks and SVMs. Furthermore, the paper proposes a new data reduction technique that is based on k-median clustering algorithm. Our experimental results illustrate that, in the case of SVMs, data reduction techniques can effectively reduce the dataset size incurring small performance degradation. In the case of Neural Networks, the performance loss is somewhat greater, for the same data reduction rate, but both SVM and Neural Network models outperform the k -NN approach that is typically used in Data Mining applications. [ABSTRACT FROM AUTHOR]

Graph sampling is a challenging problem in network analysis due to the complex structures of networks. Currently, a series of graph sampling algorithms based on random walks achieve good results in graph sampling tasks. However, the existing methods often reduce the conductance of graphs, causing the sampler to stay in the same node for a long time. This results in undersampling. In this paper, we propose a novel Weighted Jump Random Walk (WJRW) algorithm to generate representative samples. We design a parameter in the WJRW algorithm that can adjust the proportions of random walk and random jump in every step. According to the issue of repeated sample nodes in the Generalized Maximum Degree (GMD) method and the problem of large deviations in the Simple Random Walk (SRW), WJRW addresses the weaknesses of the GMD method and enhances the diffusion of a random walker on graphs, leading to a more representative sample. Then, WJRW addresses the issue of large deviations in SRW and enhances the efficiency of the unbiased estimator. By generating smoother stationary distributions. Numerical experiments with extensive real-world networks verify that our method achieves higher accuracy than classical and state-of-the-art methods in estimating distribution estimation. [ABSTRACT FROM AUTHOR]

The advancement of low-cost, non-manually labeled big data technologies for fault detection in wind turbines is crucial to guarantee their safe and efficient operation. In this context, Normal Behavior Modelling (NBM) is a prevalent approach. However, existing NBM methods for wind turbine fault detection inadequately utilize complex spatial–temporal correlations within the Supervisory Control and Data Acquisition (SCADA) during the normal behavior prediction stage. To cope with this limitation, this paper propose a hierarchical spatial-temporal autocorrelation graph neural network method for online wind turbine pitch system fault detection. This approach integrates a dynamic graph network with an adaptive structure for spatial correlation learning and a transformer model with autocorrelation attention to identify periodic time-distance correlations. Furthermore, a hierarchical neural network, specifically designed for wind turbine physical structure, enhances local information extraction through seven latent graph neural networks and a probsparse attention mechanism. The proposed method demonstrates significant effectiveness in real-world applications, as evidenced by data from a Shanghai wind farm. [ABSTRACT FROM AUTHOR]

The voluminous quantity of data accessible on social media platforms offers insight into the sentiment disposition of individual users, where multi-modal sarcasm detection is often confounding. Existing sarcasm detection methods use different information fusion methods to combine information from different modalities but ignore hidden information within modalities and inconsistent information between modalities. Discovering the implicit information within the modalities and strengthening the information interaction between modalities is still an important challenge. In this paper, we propose a Multi-Channel Enhanced Fusion (MCEF) model for cross-modal sarcasm detection to maximize the information extraction between different modalities. Specifically, text extracted from images acts as a new modality in the front-end fusion models to augment the utilization of image semantic information. Then, we propose a novel bipolar semantic attention mechanism to uncover the inconsistencies among different modal features. Furthermore, a decision-level fusion strategy from a new perspective is devised based on four models to achieve multi-channel fusion, each with a distinct focus, to leverage their advantages and mitigate the limitations. Extensive experiments demonstrate that our model surpasses current state-of-the-art models in multi-modal sarcasm detection. [ABSTRACT FROM AUTHOR]

This paper presents a comprehensive study on clustering: exiting methods and developments made at various times. Clustering is defined as an unsupervised learning where the objects are grouped on the basis of some similarity inherent among them. There are different methods for clustering the objects such as hierarchical, partitional, grid, density based and model based. The approaches used in these methods are discussed with their respective states of art and applicability. The measures of similarity as well as the evaluation criteria, which are the central components of clustering, are also presented in the paper. The applications of clustering in some fields like image segmentation, object and character recognition and data mining are highlighted. [ABSTRACT FROM AUTHOR]

Network embedding aims to learn a low-dimensional vector for each node in networks, which is effective in a variety of applications such as network reconstruction and community detection. However, the majority of the existing network embedding methods merely exploit the network structure and ignore the rich node attributes, which tend to generate sub-optimal network representation. To learn more desired network representation, diverse information of networks should be exploited. In this paper, we develop a novel deep autoencoder framework to fuse topological structure and node attributes named FSADA. We firstly design a multi-layer autoencoder which consists of multiple non-linear functions to capture and preserve the highly non-linear network structure and node attribute information. Particularly, we adopt a pre-processing procedure to pre-process the original information, which can better facilitate to extract the intrinsic correlations between topological structure and node attributes. In addition, we design an enhancement module that combines topology and node attribute similarity to construct pairwise constraints on nodes, and then a graph regularization is introduced into the framework to enhance the representation in the latent space. Our extensive experimental evaluations demonstrate the superior performance of the proposed method. [ABSTRACT FROM AUTHOR]

• We mathematically proved that several spectral alignment methods (NMF, IsoRank and FINAL) can be unified into a general heuristic form. From representation view, we further prove the heuristic is equal to a linear GCN framework without inserting self-loops. • We theoretically illustrate that there exist issues about inconsistency and attribute difference when GCN is directly applied to graph alignment. Hence, we further propose a novel Deep Graph Alignment Network (DGAN) which can effectively address this issue. • We conduct extensive experiments on public benchmarks and the experimental results show our framework notably outperforms traditional graph alignment methods by a large margin. Graph alignment, also known as network alignment has many applications in data mining tasks. It aims to find the node correspondence across disjoint graphs. Recently, various methods like representation learning methods, spectral methods have been proposed to solve the graph alignment problem, but they either only consider the local structure information but ignore the neighborhood similarity, or their alignment process is easy to be disturbed by nodes with similar structure or attribute. In this paper, we consider both center and neighborhood similarities, aiming to reduce the inconsistency between them and enlarge the difference among node representations. We further propose model DGAN(D eep G raph A lignment N etwork) containing the DNN module and GCN module to learn more unique node representations under the guidance of the attribute-supervised module. Moreover, we theoretically prove that most spectral methods can be unified into a linear GCN model. By extensive experiments on public benchmarks, we show that our model achieves a good balance between alignment accuracy and speed over multiple datasets compared with existing methods. [ABSTRACT FROM AUTHOR]

One of the main goals of Big Data research, is to find new data mining methods that are able to process large amounts of data in acceptable times. In Big Data classification, as in traditional classification, class imbalance is a common problem that must be addressed, in the case of Big Data also looking for a solution that can be applied in an acceptable execution time. In this paper we present Approx-SMOTE, a parallel implementation of the SMOTE algorithm for the Apache Spark framework. The key difference with the original SMOTE, besides parallelism, is that it uses an approximated version of k -Nearest Neighbor which makes it highly scalable. Although an implementation of SMOTE for Big Data already exists (SMOTE-BD), it uses an exact Nearest Neighbor search, which does not make it entirely scalable. Approx-SMOTE on the other hand is able to achieve up to 30 times faster run times without sacrificing the improved classification performance offered by the original SMOTE. [ABSTRACT FROM AUTHOR]

• A new method based on Locality-Sensitive Hashing (LSH) for cluster initialization in the k-means like clustering was proposed. • A new clustering algorithm called LSH-k-representatives for handling big categorical data was developed. • Extensive experiments conducted on multiple real-world and synthetic datasets have demonstrated the effectiveness of the proposed clustering method. • The newly developed algorithm achieves comparable or better clustering performance to the existing well-known k-means like algorithms, while it is significantly more efficient by a factor of between 2× and 32×. Clustering categorical data remains a challenging problem in the era of big data, due to the difficulty in measuring dis/similarity meaningfully for categorical data and the high computational complexity of existing clustering algorithms that makes it difficult to be applied in practical use for big data mining applications. In this paper, we propose an integrated approach that incorporates the Locality-Sensitive Hashing (LSH) technique into the k -means-like clustering so as to make it capable of predicting the better initial clusters for boosting clustering effectiveness. To this end, we first utilize a data-driven dissimilarity measure for categorical data to construct a family of binary hash functions that are then used to generate the initial clusters. We also propose to use a nearest neighbor search at each iteration for cluster reassignment of data objects to improve the clustering complexity. These solutions are incorporated into the k -representatives algorithm resulting in the so-called LSH- k -representatives algorithm. Extensive experiments conducted on multiple real-world and synthetic datasets have demonstrated the effectiveness of the proposed method. It is shown that the newly developed algorithm yields comparable or better clustering results in comparison to the existing closely related works, yet it is significantly more efficient by a factor of between 2 × and 32 ×. [ABSTRACT FROM AUTHOR]

Existing s tate- o f- t he- a rt (SOTA) video anomaly detection methods have mainly focused on the network design for obtaining their performance improvements. Different to the main research trend, this paper focuses on the data perspective, where the key idea is to mine useful 'normal' data from the unseen video frames. For any off-the-shelf anomaly deep model, these newly mined data could help the deep model in familiarizing more normal feature patterns. Thus, those previously miss-detected abnormal patterns and false-alarms detections would have more chances to be rectified if the target anomaly detection deep model has been finetuned on these newly mined normal data. Extensive quantitative evaluations have verified the effectiveness of the proposed approach in achieving persistent performance improvement by almost 7% AUC improvement. [ABSTRACT FROM AUTHOR]

Clustering is an important research direction in data mining. However, there is no one clustering algorithm that can be applied efficiently in all situation. Clustering ensemble is the best way to solve the above-mentioned problems. It combines the results of multiple clustering algorithms, and the final result is significantly better than a single clustering algorithm. Although there is a lot of constraint information, the existing clustering ensemble algorithm does not utilize it. This paper uses constraint information in consensus function and proposes a Semi-supervised Selective Clustering Ensemble based on Chameleon (SSCEC) and Semi-supervised Selective Clustering Ensemble based on Ncut (SSCEN) to solve the above problem. SSCEC uses the chameleon algorithm as consensus function, and processes constraint information in subgraph partition and subgraph combining. SSCEN uses the Normalized cut algorithm as consensus function, and processes constraint information in the process of graph dichotomy. The experiment results show that our proposed two semi-supervised member selection clustering ensemble algorithms are better than other semi-supervised algorithms. [ABSTRACT FROM AUTHOR]

In many data mining applications, dimensionality reduction is a primary technique to map high-dimensional data to a lower dimensional space. In order to preserve more local structure information, we propose a novel orthogonal least squares regression model for feature extraction in this paper. The main contributions of this paper are shown as follows: first, the new least squares regression method is constructed under the orthogonal constraint which can preserve more discriminant information in the subspace. Second, the optimization problem of classical least squares regression can be solved easily. However the proposed objective function is an unbalanced orthogonal procrustes problem, it is so difficult to obtain the solution that we present a novel iterative optimization algorithm to obtain the optimal solution. The last one, we also provide a proof of the convergence for our iterative algorithm. Additionally, experimental results show that we obtain a global optimal solution through our iterative algorithm even though the optimization problem is a non-convex problem. Both theoretical analysis and empirical studies demonstrate that our method can more effectively reduce the data dimensionality than conventional methods. [ABSTRACT FROM AUTHOR]

Internet of Things (IoT), an emerging computing paradigm which interconnects various ubiquitous things is facilitating the advancement of computational intelligence. This paper aims at investigating the computation intelligence extraction approach with focus on the dynamic k -clique mining that is an important issue in social network analysis. The k -clique detection problem as one of the fundamental problems in computer science, can assist us to understand the organization style and behavioral patterns of users in social networks. However, real social networks usually evolve over time and it remains a challenge to efficiently detect the k -cliques from dynamic social networks. To address this challenge, this paper proposes an efficient k -clique dynamic detection theorem based on triadic formal concept analysis (TFCA) with completed mathematical proof. With this proposed detection theorem, we prove that the k -cliques detection problem is equivalent to finding the explicit k -cliques generated from k -triadic equiconcepts plus the implicit k -cliques derived from its high-order triadic equiconcepts. Theoretical analysis and experimental results illustrate that the proposed detection algorithm is efficient for finding the k -cliques and exploring the dynamic characteristics of the sub-structures in social networks. [ABSTRACT FROM AUTHOR]

Weakly Supervised Object Localization (WSOL) is a technique for obtaining the object location from attention maps of the classification network, without using bounding box annotations. Existing WSOL approaches lack the modeling of the correlations between different regions of the target object. Hence they can only locate some discriminative attentions, which are small and sparse. Besides, they introduce too many background attentions when mining more object parts. In this paper, we propose a novel Adaptive Attention Augmentor (A3) to adaptively augment the target object attentions on class attention maps. It can supplement object attentions by discovering the semantic correspondence between different regions and dynamically suppress background attentions through the proposed Focal Dice loss. Extensive experiments demonstrate the effectiveness of our approach. On the ILSVRC dataset, A3 achieves a new state-of-the-art localization performance. On the fine-grained datasets including CUB-200–2011 and Cars-196, it also achieves very competitive results. [ABSTRACT FROM AUTHOR]

Social media is growing at an explosive rate and it becomes increasingly difficult for users to locate useful information from massive and high-velocity social media data. Recently a few social media sites have employed timelines to organize historical data of entities, which greatly improve user experiences in rediscovering important timeline episodes and understanding their order and trends. However, timelines of entities are not explicitly available in most social media sites. In other words, a gap exists between the importance of timelines and their availability in social media. In this paper, we investigate the problem of mining timelines of entities in social media. We delineate its challenges and opportunities, and propose a principled framework Timeliner, which can automatically generate timelines for entities by exploiting their historical social media data. We conduct experiments on real-world datasets, and the experimental results demonstrate that Timeliner can accurately mine timelines of entities in social media. [ABSTRACT FROM AUTHOR]

Joint classification of multi-source data is often better than single-source data in application scenes, but it is difficult to ensure effective feature extraction of multi-source information. In this paper, a similar double-concentrate network, denoted as SDCN, is proposed for extracting features effectively and classifying more accurately based on hyperspectral imagery (HSI) and Light Detection and Ranging (LiDAR) data. Specifically, the dual-concentrate network is first developed to capture spectral and spatial features from HSI, and then expanding the connection of LiDAR information based on the trained HSI branch. Each branch of the designed network is similar, which includes two convolutional layers, one maximum pooling layer, one batch normalization layer and two activation layers. After the network of HSI is fully trained, similar network is deployed to distinguish spatial features and 'band' difference of LiDAR data, and different features are also combined with multi-source associations. The absolute symmetry network structure and specific multi-source connection can ensure the orderly and balance of features extracted in this model, and adjust the direction of feature extraction constantly. Experimental results on several real data demonstrate that the proposed SDCN outperforms other relevant state-of-the-art methods. [ABSTRACT FROM AUTHOR]

With the exponential growth of the Internet of Things and Cloud Computing, especially in recent years, the potentiality of Machine Learning (ML) has been demonstrated and amplified, together with data mining developments and the availability of large amounts of data. In order to design a ML system capable of producing effective and accurate predictions and results it is necessary to make sure that it is actually working on large data sets; clean, high quality and complete data really representing the information you are trying to analyze. More data are available, the more accurate are evidently predictions. Forecasting represents an important use of extracted knowledge from data. It is the process of predict future demand for an offered product or service, and it allows for optimizing company decisions, reducing risks, managing stocks, planning sales and making many other internal or in-market assessments. In this paper we present and discuss a novel ensemble technique in forecasting workload of local health department. The proposed approach relies on a real dataset composed by over than 20 M of administrative e-health records. Obtained results demonstrate that our ensemble approach outperforms the state-of-the-art. [ABSTRACT FROM AUTHOR]

Benefited from advances of neuron tracing techniques, the ever-increasing number of digitally reconstructed 3D neuron images have greatly facilitated the research in neuromorphology. However, the sheer volume and the complexity of these 3D neuron data pose significant challenges for computational analytics, e.g., effectively finding neurons sharing similar morphologies, identifying neuron types, correlating neuron morphologies with properties, all of which require accurate measuring and fast indexing methods especially designed for the massive 3D neuronal images. In this paper, we present an accurate and efficient framework for the computational analytics of 3D neuronal structures based on advances of deep learning and data mining techniques. Particularly, unlike previous methods quantitatively describe neurons by measuring pre-defined metrics according to the tree-topological structures, we first develop a new method for the morphological feature representation by a proposed 3D neuron mapping and a modified generative adversarial networks (GANs). Subsequently, considering the computational complexity when retrieving large-scale neuron datasets, we integrate the neuron features with graph-based indexing, which can significantly improve the retrieval efficiency without losing accuracy. Experimental results show that our framework can effectively measure the similarity among massive neurons (e.g., 100 , 000 neurons), outperforming state-of-the-arts with more than 10% in accuracy and hundreds of times in efficiency improvements. [ABSTRACT FROM AUTHOR]

The self-expressiveness property of the data, i.e., each sample is linearly represented as a combination of other samples, has recently aroused much attention in the community of data mining and machine learning and shown great promising for subspace clustering. However, real-world data are usually contaminated by noise and outliers. The true similarity between samples directly learned from the original data may deviate from the intrinsic structure of the data, and subsequent clustering results will be severely affected. Hence naively taking a corrupted dictionary, i.e., data itself, will not always obtain the desired clustering performance. To address the above issues, this paper proposes a robust principal component analysis induced self-representation clustering method via adaptively learning the similarity graph from the clean data in the self-representation framework, where the original data are recovered, and the representation coefficients are obtained simultaneously in a unified framework. Specifically, we jointly integrate a nonnegative constraint and a distance regularization into the proposed framework, which guarantees the learned affinity matrix can simultaneously capture the global and local structure of data. Experimental results on both synthetic data and several famous multimedia datasets demonstrate that the proposed method performs much better against state-of-the-arts. [ABSTRACT FROM AUTHOR]

Proactively and naturally guiding the dialog from the non-recommendation context (e.g., Chit-chat) to the recommendation scenario (e.g., Music) is crucial for the Conversational Recommender System (CRS). Prior studies mainly focus on planning the next dialog goal (e.g., chat on a movie star) conditioned on the previous dialog. However, we find the dialog goals can be simultaneously observed at different levels, which can be utilized to improve CRS. In this paper, we propose D ual-space H ierarchical L earning (DHL) to leverage multi-level goal sequences and their hierarchical relationships for conversational recommendation. Specifically, we exploit multi-level goal sequences from both the representation space and the optimization space. In the representation space, we propose the hierarchical representation learning where a cross attention module derives mutually enhanced multi-level goal representations. In the optimization space, we devise the hierarchical weight learning to reweight lower-level goal sequences, and introduce bi-level optimization for stable update. Additionally, we propose a soft labeling strategy to guide optimization gradually. Experiments on two real-world datasets verify the effectiveness of our approach. Code and data are available here. [ABSTRACT FROM AUTHOR]

This paper introduces a novel system for information extraction from visually rich documents (VRD) using a weighted graph representation. The proposed method aims to improve the performance of the information extraction task by capturing the relationships between various VRD components. The VRD is modeled as a weighted graph, in which visual, textual, and spatial features of text regions are encoded in nodes and edges representing the relationships between neighboring text regions. The information extraction task from VRD is performed as a node classification task through the use of a graph convolutional networks, where the VRD graphs are fed into the network. The approach is evaluated across diverse documents, encompassing invoices and receipts, revealing achievement levels equal to or surpassing robust baselines. [Display omitted] • Information Extraction. • Visually Rich Documents. • Graph Convolutional Networks. [ABSTRACT FROM AUTHOR]

Official gazettes are documents published by governments to publicize their actions, spanning long periods of time and making an important transparency mechanism. These documents have information on laws, contracts, and bidding processes, as well as on civil servants and their careers in public service. Automatic information extraction of these documents may contribute to public transparency, with two tasks being especially useful: the classification of the different segments of these documents, the so called acts; and the Named Entity Recognition (NER) within the acts. The variety of official gazettes and their patterns brings up the necessity of constructing different tools for specific gazettes. In this paper, we propose DODFMiner, a command-line interface tool to classify acts and extract named entities from the Official Gazette of the Federal District. The tool follows a 3-step approach: the pre-processing of the input data; text classification using rule-based systems with regular expressions; and NER with Machine Learning algorithms. It allows users to input JSON files and receive CSV as output, providing information that allows users to track government procurements through years, contracts duration and total amount, among others. We also propose a set of experiments to support the choice of models included in the tool, covering the classification and NER steps. Text classification achieved a mean F1-score of 0.778, while to the NER, we compared 3 different architectures, CRF with a mean F1-score of 0.851, CNN-biLSTM-CRF with 0.787 and CNN-CNN-LSTM with 0.841. [ABSTRACT FROM AUTHOR]

Feature selection is a vital preprocessing step in real applications of data mining and machine learning. With the prevalence of high-dimensional hybrid data sets in real-world scenarios, along with the presence of test costs and misclassification costs, the need for effective feature selection methods has become more prominent. However, existing feature selection approaches mainly focus on cost-sensitive data from a single granularity perspective, and they are primarily applicable to single-typed data sets. To address these limitations, this paper presents a novel feature selection approach specifically designed for hybrid data, considering variable test costs and misclassification costs. The proposed method is based on neighborhood multigranulation rough sets (NMRS), which provides more comprehensive and multi-angle data analysis for cost-sensitive hybrid data. First, a novel multigranulation model is developed to effectively process cost-sensitive hybrid data. Building upon this model, a cost-based multi-criteria measure is proposed to evaluate the significance of features. This measure takes into account the comprehensive information of candidate features, including their power in algebraic view, information view, and associated costs. Furthermore, a heuristic feature selection algorithm based on NMRS is proposed to handle hybrid data with the test costs and misclassification costs. This algorithm leverages the benefits of the proposed multigranulation model and cost-based measure to identify the most discriminative features efficiently. Finally, the experimental results on twelve different datasets show that the proposed heuristic algorithm outperforms other compared algorithms, especially in total cost and classification accuracy. [ABSTRACT FROM AUTHOR]

• Context learning extraction from ship trajectory data. • AIS data real-world data use for data mining problems and anomaly detection. • Trajectory compression and segmentation techniques. • Data mining techniques for trajectory clustering applications. This paper presents a context information extraction process over Automatic Identification System (AIS) real-world ship data, building a system with the capability to extract representative points of a trajectory cluster. With the trajectory cluster, the study proposes the use of trajectory segmentation algorithms to extract representative points of each trajectory and then use the k-means algorithm to obtain a series of centroids over all the representative points. These centroids, combined, form a new representative trajectory of the cluster. This new representative trajectory of the input cluster represents new contextual information extracted from the original set of trajectories, being possible to apply anomaly detection approaches over the new obtained context. The results show a suitable approach with several compression algorithms that are compared with a metric based on the Perpendicular Euclidean Distance. [ABSTRACT FROM AUTHOR]

A huge number of videos are posted every day on social media platforms such as Facebook and YouTube. This makes the Internet an unlimited source of information. In the coming decades, coping with such information and mining useful knowledge from it will be an increasingly difficult task. In this paper, we propose a novel methodology for multimodal sentiment analysis, which consists in harvesting sentiments from Web videos by demonstrating a model that uses audio, visual and textual modalities as sources of information. We used both feature- and decision-level fusion methods to merge affective information extracted from multiple modalities. A thorough comparison with existing works in this area is carried out throughout the paper, which demonstrates the novelty of our approach. Preliminary comparative experiments with the YouTube dataset show that the proposed multimodal system achieves an accuracy of nearly 80%, outperforming all state-of-the-art systems by more than 20%. [ABSTRACT FROM AUTHOR]

Multiple graph clustering is an important tool in data integration and data mining for graph-based data. The prediction and classification accuracy can be significantly improved by integrating information from multiple sources and data sets. The aim of multiple graph clustering is to partition objects into several clusters such that clusters in each graph are well-separated and clusters across different graphs are consistent. Existing methods assume that the degrees of association among different graphs are the same. However, some graphs may be strongly or weakly associated with the other graphs due to high or low correlation of their associated cluster structures. When their cluster structures are (or are not) similar, their degree of association should be high (or low). Accurate clustering results can be obtained by integrating such association information in multiple graphs. The main aim of this paper is to study multiple graph semi-supervised clustering by considering a large amount of multiple graph data and a small amount of labeled data. We propose a constrained optimization problem that can determine cluster structures and degrees of association simultaneously in multiple graph clustering. In our formulation, we make use of orthogonality for cluster structure indicator and cosine correlation for degree of association. With orthogonality constraint in the clustering process, we develop a gradient flow method to solve the resulting optimization problem. And the convergence of the proposed iterative method is also shown. Numerical examples including synthetic data and real data sets with few known labels are tested, and are presented to show the efficiency and effectiveness of our proposed method compared with the testing methods in the literature. [ABSTRACT FROM AUTHOR]

Weakly supervised semantic segmentation has been attracting increasing attention as it can alleviate the need for expensive pixel-level annotations through the use of image-level labels. Relevant methods mainly rely on the implicit object localization ability of convolutional neural networks (CNNs). However, generated object attention maps remain mostly small and incomplete. In this paper, we propose an Atrous Convolutional Feature Network (ACFN) to generate dense object attention maps. This is achieved by enhancing the context representation of image classification CNNs. More specifically, cascaded atrous convolutions are used in the middle layers to retain sufficient spatial details, and pyramidal atrous convolutions are used in the last convolutional layers to provide multi-scale context information for the extraction of object attention maps. Moreover, we propose an attentive fusion strategy to adaptively fuse the multi-scale features. Our method shows improvements over existing methods on both the PASCAL VOC 2012 and MS COCO datasets, achieving state-of-the-art performance. [ABSTRACT FROM AUTHOR]

Data often exists in subspaces embedded within a high-dimensional space. Subspace clustering seeks to group data according to the dimensions relevant to each subspace. This requires the estimation of subspaces as well as the clustering of data. Subspace clustering becomes increasingly challenging in high dimensional spaces due to the curse of dimensionality which affects reliable estimations of distances and density. Recently, another aspect of high-dimensional spaces has been observed, known as the hubness phenomenon, whereby few data points appear frequently as nearest neighbors of the rest of the data. The distribution of neighbor occurrences becomes skewed with increasing intrinsic dimensionality of the data, and few points with high neighbor occurrences emerge as hubs. Hubs exhibit useful geometric properties and have been leveraged for clustering data in the full-dimensional space. In this paper, we study hubs in the context of subspace clustering. We present new characterizations of hubs in relation to subspaces, and design graph-based meta-features to identify a subset of hubs which are well fit to serve as seeds for the discovery of local latent subspaces and clusters. We propose and evaluate a hubness-driven algorithm to find subspace clusters, and show that our approach is superior to the baselines, and is competitive against state-of-the-art subspace clustering methods. We also identify the data characteristics that make hubs suitable for subspace clustering. Such characterization gives valuable guidelines to data mining practitioners. [ABSTRACT FROM AUTHOR]

The acceleratinnsional nodal attributes in various data mining tasks highlights the significance of feature selection on the networked data. Due to the lack of class labels of nodes, many feature selection methods are proposed in semi-supervised or unsupervised manners in various scenarios instead of supervised ones. More often than not, features and (pseudo) labels are correlated in a nonlinear way that is more intricate than linearity. In these circumstances, the vast majority of existing linear algorithms could not work well since they select features according to how well the feature can linearly explain the variance of labels. Moreover, although some methods focus on nonlinear feature selection, with the neglect of the link relations between data, they are difficult to be applied to attributed networks directly. In this paper, we investigate how to achieve nonlinear feature selection on attributed networks with the help of both labeled and unlabeled data. Methodologically, we first propose a novel semi-supervised nonlinear framework FS-GCN based on graph convolutional networks (GCNs) to select high-quality features, which can elaborately catch the nonlinear dependency between nodal attributes and class labels. To verify the importance of nonlinearity precisely, we further explore the possibility of totally removing the label information so that a variant of FS-GCN is proposed in the unsupervised form, referred to as UFS-GCN. Besides, experimental results on several real-world datasets validate the superiority of FS-GCN as well as UFS-GCN in terms of the quality of selected features, suggesting their robustness in the condition of extremely low even zero label ratio. [ABSTRACT FROM AUTHOR]