Search

Exam paper generation is an indispensable part of teaching. Existing methods focus on the use of question extraction algorithms with labels for each question provided. Obviously, manual labeling is inefficient and cannot avoid label bias. Furthermore, the quality of the exam papers generated by the existing methods is not guaranteed. To address these problems, we propose a novel approach to generating exam papers based on prediction of exam performance. As such, we update the quality of the initially generated questions one by using dynamic programming, as well as in batches by using genetic algorithms. We performed the prediction task by using Deep Knowledge Tracing. Our approach considered the skill weight, difficulty, and distribution of exam scores. By comparisons, experimental results indicate that our approach performed better than the two baselines. Furthermore, it can generate exam papers with adaptive difficulties closely to the expected levels, and the related student exam scores will be guaranteed to be relatively reasonable distribution. In addition, our approach was evaluated in a real learning scenarios and shows advantages.

Surrogate-assisted evolutionary algorithms (SAEAs), which use efficient surrogate models or meta-models to approximate the fitness function in evolutionary algorithms (EAs), are effective and popular methods for solving computationally expensive optimization problems. During the past decades, a number of SAEAs have been proposed by combining different surrogate models and EAs. This paper dedicates to providing a more systematical review and comprehensive empirical study of surrogate models used in single-objective SAEAs. A new taxonomy of surrogate models in SAEAs for single-objective optimization is introduced in this paper. Surrogate models are classified into two major categories: absolute fitness models, which directly approximate the fitness function values of candidate solutions, and relative fitness models, which estimates the relative rank or preference of candidates rather than their fitness values. Then, the characteristics of different models are analyzed and compared by conducting a series of experiments in terms of time complexity (execution time), model accuracy, parameter influence, and the overall performance when used in EAs. The empirical results are helpful for researchers to select suitable surrogate models when designing SAEAs. Open research questions and future work are discussed at the end of the paper.

Experton theory, a generalization of probabilistic set theory, that is of great usefulness to group decision analysis, was first proposed as a means of improving the processing and analysis of opinions issued by experts. This theory produces an information-fusion mathematical object , the experton, which can be used in predictive problems to justify decisions based on well-constructed reasoning. The aim of this paper is to present an aggregative method of several expertons, with the idea that some of the groups of experts involved in producing these expertons may have more influence than others in the decision-making process. In this article, we carry out an aggregation analysis of expertons, not experts, which culminates in the creation of a new mathematical object. This object, which is called the weighted average multiexperton, is coherent with an experton-type object created from a weighting of the initial data provided by all experts. Since the aggregation method presented has been devised to represent the decision-maker’s attitude regarding the importance of different groups of experts, this approach represents a new tool for dealing with group decision-making problems. Additionally, the study presents some properties of the new object. Finally, the paper ends with an application for business decision-making.

In association rules mining from data that have numeric-valued attributes, automatically adjusting the attribute intervals at the time of the mining process without a preprocess is very critical for preventing data loss and attribute interactions. In this paper, differential evolution and sine cosine algorithm based novel hybrid multi-objective evolutionary optimization methods are proposed for rapidly and directly mining the reduced high-quality numerical association rules by simultaneously adjusting the relevant intervals of related attributes without finding the frequent itemsets. These algorithms perform a global search and find the high-quality rules set in only one execution by modeling the rule mining task as a multi-objective problem that simultaneously meets different conflicting metrics. The algorithms proposed in this paper ensure the discovered rules to have high confidence and support and to be comprehensible. The proposed methods automate the rule mining process by directly finding the minimum intervals for the attributes and eliminating the need for minimum confidence and minimum support determined beforehand for each data set. The performances of new algorithms proposed in this study were tested with those of the state-of-the-art algorithms. The results show superiority of the proposed methods on the data sets that contain fewer attributes and higher number of instances.

Acute hypotensive episode (AHE) is a common serious postoperative complication in ICU, which may raise multiple system failure (especially of cardiac and respiratory kinds), and even cause death. Timely and effective clinical intervention is obviously vital to the saving of patients. AHE detection involves physiological time-series monitoring, processing and prediction technologies, which can offer insights to neuroscientists, biologists, and even provide support for clinicians. This paper presents a hybrid artificial intelligence model combined with CEEMDAN (complete ensemble empirical mode decomposition with adaptive noise, a typical method for physiological signal decomposition), deep learning, multiple gene expression programming and fuzzy expert system for AHE detection. In this paper, the physiological signal is selected from a benchmark dataset, for example MIMIC-II (Multiparameter Intelligent Monitoring in Intensive Care II), which collects large scale real patients’ data for clinical research. In the hybrid model, a typical signal decomposition method is employed for AHE signal processing, and an autoencoder based deep neural network is established for feature extraction. Finally, a reliable and explainable classifier is presented by fusing gene expression programming and the fuzzy method. Experimental results based on real data set demonstrate that the proposed method outperforms state-of-the-art AHE detection methods by achieving the prediction accuracy of 88.14% in 2866 records.

Highlights • The minimum cost consensus model is proposed to coordinate the opinions of experts. • A new method of LSGDM clustering is proposed to improve CRP efficiency. • A consensus index CI reflecting the harmony degree between experts is devised. • A robust counterpart of robust optimization consensus is implemented., Recently, large-scale group decision making (LSGDM) in social network comes into being. In the practical consensus of LSGDM, the unit adjustment cost of experts is difficult to obtain and may be uncertain. Therefore, the purpose of this paper is to propose a consensus model based on robust optimization. This paper focuses on LSGDM, considering the social relationship between experts. In the presented model, an expert clustering method, combining trust degree and relationship strength, is used to classify experts with similar opinions into subgroups. A consensus index, reflecting the harmony degree between experts, is devised to measure the consensus level among experts. Then, a minimum cost model based on robust optimization is proposed to solve the robust optimization consensus problem. Subsequently, a detailed consensus feedback adjustment is presented. Finally, a case study and comparative analysis are provided to verify the validity and advantage of the proposed method.

Considering that most popular models solving aspect-level sentiment classification problems mainly focus on designing complicated neural networks to scale the importance of each word in the sentence, this paper addresses this problem from the view of semantic space. Motivated by the fact that the senses of a word can be sophisticatedly embedded into the semantic space using a distributed representation, this paper hypothesizes that each sense of a word can be represented by one or more specific dimensions, and thus the target of aspect-level sentiment classification can be simplified into searching the related dimensions for the aspects and sentiments concerned. Particularly, an Attention Vector (ATV) based on attention mechanism is designed for each aspect in terms of a specific task, which involves two sub-vectors, i.e., a Dimension Attention Vector (DATV) and a Sentiment Attention Vector (SATV). The DATV determines the significances of different dimensions based on their correlations with an aspect; and the SATV allocates weights for the attributes of words, which are decided by sentiment polarities and part-of-speech (PoS) tagging. Given a sub-dataset related to an aspect, the ATV will be optimized by an Artificial Bee Colony (ABC) algorithm with a Support Vector Machine (SVM) classifier, the objective of which is to maximize classification accuracy. Intrinsically, the DATV can reduce the ambiguity existing in polysemy, meanwhile, the SATV is an auxiliary means for the optimization of the DATV, which can help eliminate the misunderstandings caused by antonyms. Then, the optimized DATV will be applied on a Convolutional Neural Network (CNN) model via simply scaling the pretrained word embeddings as inputs (named as ATV-CNN model). Experimental results show that the ATV-CNN model can have substantial advantages when compared with state-of-the-art models.

The comprehensive logistics distribution center location selection (CLDCLS) problem is a multiattribute group decision-making (MAGDM) problem in which multiple commodity preference weights are considered. To better describe the preference information and expert evaluation information, this paper utilizes 2-dimensional linguistic (2DL) information to express the preference information of various commodities and the expert evaluation, which can represent not only the evaluation information of experts but also the reliability of the evaluation information. Additionally, for solving the CLDCLS problem, this paper puts forward improved operational rules a score function, a distance formula and a correlation coefficient measure. Based on the 2DL information and the improved operational rules, we propose a 2-dimensional linguistic similarity-degree-based clustering analysis method, the 2-dimensional linguistic partitioned Maclaurin symmetric mean (2DLPMSM) operator, and the 2-dimensional linguistic weighted partitioned Maclaurin symmetric mean (2DLWPMSM) operator. The corresponding properties and special cases are demonstrated. By using these proposed methods, this paper constructs a MAGDM solution framework for the CLDCLS problem. A practical case of the CLDCLS problem is presented to demonstrate the effectiveness, rationality, robustness and superior performance of the proposed method.

Bot detection is a fundamental and crucial task for tracing and mitigating cyber threats in the Internet. This paper aims to address two major limitations of current bot detection systems. First, existing flow-based bot detection approaches ignore structural information of botnets, which lead to false detection. Second, they cannot identify the interactive behavioral patterns among heterogeneous botnet objects. In this paper, we propose a novel bot detection framework, namely Bot-AHGCN, which models fine-grained network flow objects (e.g., IP, response) as a multi-attributed heterogeneous graph and transforms bot detection problem into a semi-supervised node classification task on the graph. Particularly, we first build a multi-attributed heterogeneous information network (AHIN) to model the interdependent relationships among botnet objects. Second, we present a weight-learning based node embedding method, which learns the interactive behavioral patterns among bots and integrates them into weighted similarity graphs. Finally, we perform graph convolution on the learned similarity graphs to characterize more comprehensive and discriminative features of bots, and feed them into a forward neural network to identify bots. The overall experimental results on two real-world datasets confirm that Bot-AHGCN outperforms the existing state-of-the-art approaches, and presents better interpretability by introducing meaningful meta-paths and meta-graphs.

Association Rule Mining (ARM) is a significant task for discovering frequent patterns in data mining. It has achieved great success in a plethora of applications such as market basket, computer networks, recommendation systems, and healthcare. In the past few years, evolutionary computation-based ARM has emerged as one of the most popular research areas for addressing the high computation time of traditional ARM. Although numerous papers have been published, there is no comprehensive analysis of existing evolutionary ARM methodologies. In this paper, we review emerging research of evolutionary computation for ARM. We discuss the applications on evolutionary computations for different types of ARM approaches including numerical rules, fuzzy rules, high-utility itemsets, class association rules, and rare association rules. Evolutionary ARM algorithms were classified into four main groups in terms of the evolutionary approach, including evolution-based, swarm intelligence-based, physics-inspired, and hybrid approaches. Furthermore, we discuss the remaining challenges of evolutionary ARM and discuss its applications and future topics.

In the smart grid, efficient power supplies require near-real-time users’ electricity usage metering data, but these data might leak users’ private information, e.g., living habits. To address this problem, a number of privacy-preserving data aggregation schemes have been proposed in the literature. In this paper, we present a new type of attack, called malicious data mining attack, by which the adversary can infer a target user’s electricity usage data. When considering this attack, the majority of existing data aggregation schemes have one of the following two shortcomings. In one aspect, the schemes based on homomorphic encryption can output an accurate aggregation result, but most of them are vulnerable to this attack. In another aspect, the schemes based on differential privacy able to withstand this attack, but the random noises introduced prevent accurate aggregation results from being computed. In this paper, we propose a novel data aggregation scheme that is not only secure against the malicious data mining attack, but also capable of outputting an accurate aggregation result. Detailed security analyses indicate that the proposed scheme satisfies the desirable properties for privacy-preserving data aggregation in the smart grid, and the simulated results demonstrate that our proposed scheme enjoys low computation and communication overhead.

Nowadays, computer science has the ability of assisting experts in different application areas. Recently there has been increasing attention in the health area, and there exist different approaches based on artificial intelligence that have been proposed in the diagnosis of several kinds of diseases. In particular, fuzzy systems have been successfully used as Diagnosis Systems, in this way helping doctors to realize a faster and more accurate diagnosis. However, with the emergence of Type-2 Fuzzy Systems, there have been important improvements in handling the uncertainty with respect to traditional Fuzzy Systems (now called Type-1 Fuzzy Systems) in different kinds of problems. In the present paper, a new approach to Fuzzy Diagnosis based on Type-2 Fuzzy Systems is proposed and compared with respect to Type-1 Fuzzy Systems on a set of diagnosis problems, in order to evaluate the relevance of the uncertainty handling in this kind of problems. On the other hand, the paper is also aiming at observing the accuracy behavior in Fuzzy Diagnosis Systems by changing the uncertainty level in the models. Finally, a comparison of Interval Type-2 Fuzzy Systems with respect to General Type-2 Fuzzy Systems for a set of diagnosis problems is presented.

Cyber code analysis is fundamental to malware detection and vulnerability discovery for defending cyber attacks. Traditional approaches resorting to manually defined rules are gradually replaced by automated approaches empowered by machine learning. This revolution is accelerated by big code from open source projects which support machine learning models with outstanding performance. In the context of a data-driven paradigm, this paper reviews recent analytic research on cyber code of malicious and common software by using a set of common concepts of similarity, correlation and collective indication. Sharing security goals in recognizing anomalous code that may be malicious or vulnerable. The ability to do so is not determined in isolation, rather drawn for code correlation and context awareness. This paper demonstrates a new research methodology of data driven cyber security (DDCS) and its application in cyber code analysis. The framework of the DDCS methodology consists of three components, i.e., cyber security data processing, cyber security feature engineering, and cyber security modeling. Some challenging issues are suggested to direct the future research.

In the field of mobile user behavior analysis, clustering algorithm is used to do the user classification . Usually the mobile user’s dataset is mixed, which contains both numerical and categorical type of data. It leads to inaccurate results when doing the user classification using traditional algorithms like K-Means and which are affected by the initialization process extremely. On the other hand, K-Prototypes algorithm is used to process the mixed data. It is difficult to ascertain the coefficient of classification attribute weight. Based on the above problems, this paper proposes a mixed attributes oriented dynamic SOM fuzzy cluster algorithm (D-SOMFCM-OMA) for mobile user classification. Firstly, the algorithm proposed in this paper gives the primary clustering using Self-Organizing feature Map (SOM) to get the initial clustering parameters. As the preprocessing of clustering, this step reduces effects caused by inappropriate initialization. Then, the algorithm utilizes the improved dynamic fuzzy K-Prototypes cluster method in the second-time clustering to classify users dynamically. It calculates weight of all kinds of attributes according to the proportion of the attribute and uses fine-tuned coefficient to adjust the attribute weight. For improving the clustering effect, the algorithm uses Jaccard distance to calculate the distance in the mixed attribute variables. Further, this paper defines the user mean membership threshold which is an indicator to determine whether different groups need to be added. Finally, some comparison experiments are conducted on the UCI standard datasets to show the advantage of the improved fuzzy clustering algorithm oriented the mixed attributes (IFCM-OMA). Also the other experiment using dataset of UCI verify the validity of the algorithm of D-SOMFCM-OMA.

Class imbalanced datasets are common across different domains including health, security, banking and others. A typical supervised learning algorithm tends to be biased towards the majority class when dealing with imbalanced datasets. The learning task becomes more challenging when there is also an overlap of instances from different classes. In this paper, we propose an undersampling framework for handling class imbalance in binary datasets by removing potential overlapped data points. Our methods are designed to identify and eliminate majority class instances from the overlapping region. Accurate identification and elimination of these instances maximise the visibility of the minority class instances and at the same time minimises excessive elimination of data, which reduces information loss. Four methods based on neighbourhood searching with different criteria to identify potential overlapped instances are proposed in this paper. Extensive experiments using simulated and real-world datasets were carried out. Results show comparable performance with state-of-the-art methods across different common metrics with exceptional and statistically significant improvements in sensitivity.

A kernel-based one-class classifier is mainly used for outlier or novelty detection. Kernel ridge regression (KRR) based methods have received quite a lot of attention in recent years due to its non-iterative approach of learning. In this paper, KRR-based one-class classifier (KOC) has been extended for learning using privileged information (LUPI) framework. LUPI-based KOC method is referred to as KOC+ in this paper. This privileged information is available as feature/features of the dataset, but only during training (not during testing). KOC+ utilizes privileged features information differently compared to other features information. It uses this information in KOC+ by the help of so-called correction function . This information helps KOC+ in achieving better generalization performance . Existing and proposed classifiers are evaluated on the datasets taken from UCI machine learning repository and MNIST dataset. Moreover, experimental results exhibit that KOC+ outperforms KOC and other LUPI-based state-of-the-art one-class classifiers. Source code of this paper is provided on the corresponding author’s GitHub homepage: https://github.com/Chandan-IITI/KOCPlus_or_OCKELMPlus_or_OCLSSVMPlus

Internet of Things (IoT) is expanding rapidly and so is the number of devices, sensors and actuators joining this world. IoT devices are an important part of the data collection process in Big Data systems, so by protecting them we support and improve the security of the whole system. ZigBee is a secure communication system for the underlying Internet of Things (IoT) infrastructure. Even though ZigBee has a strong security stack built on a variety of secret keys, ZigBee devices are vulnerable to the side-channel and key extraction attacks. Due to the low cost and limited resources, most ZigBee devices store their secret keys in plaintext. In this paper, we focus on protecting the storage of ZigBee secret keys and show how Physical Unclonable Functions (PUFs) can help the ZigBee devices to be robust tamper-resistant against the physical attacks. The proposed schemes include PUF-based key storage protection and key generation. The experiments in this paper were done using SRAM-PUF. Furthermore, two algorithms were proposed to overcome the defects in the randomness of keys generated using SRAM-PUF and, at the same time, to increase the reliability of these keys. We were able to significantly improve the hardware security of ZEDs by protecting their keying materials using costless, high secure, random, stable and volatile PUF-based secret keys.

This paper proposes a novel top-down approach to rule-based fuzzy systems, one that begins with the product—an equation—and then addresses the unique features of the product, without requiring the reader to know anything about fuzzy sets and systems. The "products" are adaptive variable-structure basis function expansions , where "adaptive variable-structure" means that different subsets of its basis functions are active (non-zero) in different regions of the state space, something that occurs automatically by virtue of the structure of the basis functions, so that the products can be said to "adapt" to locations in the state space. These products are novel candidates for machine learning . Unique features of all products are: (1) Number of basis functions is no longer a variable and is established locally through type-1 or type-2 uncertainty partitioning of each variable; (2) Both coarse and fine sculpting of the state space are achieved, and are described in terms of first- and second-order partitions of the state space, respectively; and (3) Linguistic interpretability is obtained, which may be of value to an end-user. Learning about rule-based fuzzy systems can be greatly compressed by using the top-down approach that is described in this paper.

The big data samples are important source for analytics. However, its relevant/irrelevant information, unspecified variables/scales, noise/null, and so forth impose huge challenges on the analysis of relevance, feature, cause, and evaluation. This paper proposes an evidential analytics to disclose buried information in big data samples. Technically, it models memberships composed of relevant preference and replaces data with these priors. Its operations include generating analytics baselines, reducing variables, identifying sparse features, and inducing rules by taking advantage of evidence. In illustration, a case study of semiconductor manufacturing in UCI secom is presented. It discloses relevant signals, key factors, variables’ thresholds, sparse characteristics, and causal effect of damages buried in normal samples. The contribution of this paper not only contains these achievements but provides priori data for inference.

The choice of the most promising algorithm to model and predict a particular phenomenon is one of the most prominent activities of the temporal data forecasting. Forecasting (or prediction), similarly to other data mining tasks , uses empirical evidence to select the most suitable model for a problem at hand since no modeling method can be considered as the best. However, according to our systematic literature review of the last decade, few scientific publications rigorously expose the benefits and limitations of the most popular algorithms for time series prediction. At the same time, there is a limited performance record of these models when applied to complex and highly nonlinear data. In this paper, we present one of the most extensive, impartial and comprehensible experimental evaluations ever done in the time series prediction field. From 95 datasets, we evaluate eleven predictors, seven parametric and four non-parametric, employing two multi-step-ahead projection strategies and four performance evaluation measures. We report many lessons learned and recommendations concerning the advantages, drawbacks, and the best conditions for the use of each model. The results show that SARIMA is the only statistical method able to outperform, but without a statistical difference, the following machine learning algorithms : ANN , SVM , and k NN-TSPI. However, such forecasting accuracy comes at the expense of a larger number of parameters. The evaluated datasets, as well detailed results achieved by different indexes as MSE , Theil’s U coefficient, POCID, and a recently-proposed multi-criteria performance measure are available online in our repository. Such repository is another contribution of this paper since other researchers can replicate our results and evaluate their methods more rigorously. The findings of this study will impact further research on this topic since they provide a broad insight into models selection, parameters setting, evaluation measures, and experimental setup.

With the continued growth of interoperable software developed for Internet of Things (IoT), there is a growing demand to predict software defect at various testing and operational phases. This paper solves the software defect prediction problem by proposing a novel model with the help of a local tangent space alignment support vector machine (LTSA-SVM) algorithm. The model employs the SVM algorithm as the basic classifier of software defect distribution prediction model. Then, the model parameters are optimized by combining a grid search method and ten-fold cross validation. In the traditional dimensionality reduction algorithms, data loss caused by the poor attributes of data nonlinearity reduces the accuracy of SVM. Aiming at this problem, this paper uses a LTSA algorithm to extract the intrinsic structure of low-dimensional data and performs effective dimension reduction. The SVM algorithm is trained by the reduced dimension data. Finally, the feasibility of the prediction model is verified. Compared with the single SVM and the LLE-SVM prediction algorithm, the prediction model in this paper improves the prediction accuracy and F-measure by 1–4%.

This paper provides a new methodology for calibrating the fuzzy sets that are used in fsQCA, one that is based on clearly distinguishing between a linguistic variable and the linguistic terms for that variable, and that allows for uncertainties about those terms to be included in the calibration method. Each resulting fuzzy set, called an approximated reduced-information level 2 fuzzy set (RI L2 fuzzy set), is equivalent to a standard type-1 fuzzy set, but is for the linguistic variable, and, it has an S-shape, the kind of shape that is so widely used by fsQCA scholars, and is so important to fsQCA. This new calibration methodology is applied to Ragin's Breakdown of Democracy example, using new data provided by him, and demonstrates that his earlier solutions are also obtained using our approximated RI L2 fuzzy sets, something that should be reassuring to fsQCA scholars. Additionally, because the S-shaped membership functions are derived from footprints of uncertainty for all of the linguistic variable's terms, this paper shows how to obtain more precise statements of fsQCA causal combinations for their best instances, something that may be of added value to practitioners of fsQCA. Finally, we explain how different data-driven calibration robustness studies can be performed, something that may also be of great value to fsQCA practitioners.

Belief rules extend traditional IF-THEN rules to represent vagueness, incompleteness, and nonlinear causal relationships by assigning belief degrees to singletons or the universe of all possible values that the consequents of rules can take. First, this paper extends belief rules by assigning belief degrees to the subsets of all possible values that the consequents of rules with interval-valued rule weights can take. Then, this paper proposes a Valuation-Based System (VBS) approach for the modeling and risk assessment of extended belief rule-based expert systems. Finally, the proposed VBS approach is applied to two use cases for evaluating the occurrence probabilities of accidents: one is a car equipped with Automated Speed Control (ASC) using values from experts, and the other is hazardous material (hazmat) transportation accidents using real statistical data.

Experts are often solicited to provide their opinions on systems unavailable inputs. In certain cases, we can have several opinions for each system input. In such situation, we ask ourselves what is the best way to combine these opinions? For each input, we can either combine them, before their propagation into the system model, or combine them after the propagation of each opinions combination separately. The purpose of this paper is to explore the differences between these two aggregation modes. For our reliability model, outcomes show that the location of the operating point in the probabilistic space and the divergence (gap) between expert opinions are the main factors explaining the difference between both aggregation modes. In this paper, we propose the Divergence Metric δ to measure the divergence between experts’ opinions and we suggest the use of the ‘Cumulative Distribution Averaging’ as an aggregation rule. This rule seems suitable for probabilistic and non-probabilistic opinions and it avoids limitations encountered with expert opinions expressed according to bounded distributions.

During the past decade, graph-based learning methods have proved to be an effective tool to make full use of both labeled and unlabeled data samples to improve learning performance. These methods try to discover the intrinsic structures and discriminative information embedded in the data, by building one or more graphs to model the relationship among the data samples. Consequently, how to build an effective graph is the core problem. In this paper we introduce a novel graph-based classification method, called Supervised clustering-based Regularized Least Squares Classification (SuperRLSC), in which local and global graphs of the data are built by supervised clustering. The motivation is that supervised clustering may discover more actual data structures compared to unsupervised clustering. In our algorithm, we firstly employ supervised k-means to partition the whole training dataset into several meaningful clusters in order to discover the intrinsic and discriminative structures. We then use the discovered structures to build local and global graphs of the data. The local graph reveals local geometric and discriminative structures, while the global graph reveals global discriminative information. Finally a hybrid local/global graph-based regularization term is embedded into supervised classification (i.e., RLSC in this paper). To validate the effectiveness of our algorithm, a series of experiments are performed on several UCI benchmark datasets. The results show that our algorithm can achieve better or at least comparable performance to the other graph-based algorithms and the traditional state-of-the-art supervised classification methods.

Social networking services allow users to adopt and spread information via diffusion actions, e.g., share, retweet, and reply. Real applications such as viral marketing and trending topic detection rely on information diffusion. Given past items with diffusion records on a social network, this paper aims at forecasting who will participate in the diffusion of a new item c (we use hashtags in the paper) with its k earliest adopters, without using content and profile information, i.e., finding which users will adopt c in the future. We define the Diffusion Participation Forecasting (DPF) problem, which is challenging since all users except for early adopters can be the candidates, comparing to existing studies that predict which one-layer followers will adopt a new hashtag given past diffusion observations with content and profile info. To solve the DFP problem, we propose an Adoption-based Participation Ranking (APR) model, which aims to rank the actual participants in reality at higher positions. The first is to estimate the adoption probability of a new hashtag for each user while the second is a random walk-based model that incorporates nodes with higher adoption probability values and early adopters to generate the forecasted participants. Experiments conducted on Twitter exhibit that our model can significantly outperform several competing methods in terms of Precision and Recall. Moreover, we demonstrate that an accurate DPF can be applied for effective targeted marketing using influence maximization and boosting the accuracy of popularity prediction in social media.

The paper deals with two crucial steps in multi-criteria decision analysis, that are consistency of the judgments and priority vectors for alternatives/criteria. From a side, several consistency indices are proposed for measuring the consistency of a Pairwise Comparison Matrix. From another one, conditions weaker than consistency, such as transitivity and weak consistency, are proposed for representing further levels of coherence of a Decision Maker when he/she expresses his/her preferences by means of reciprocal Pairwise Comparison Matrix. Firstly, in this paper, a simulation is performed in order to establish a relation between random consistency index and coherence level. Then, since weak consistency ensures reliability to priority vectors proposed in literature, a second simulation is performed in order to measure, in case of no weak consistency, the reliability of these priority vectors.

Pawlak’s rough membership functions not only give numerical characterizations of Pawlak’s rough set approximations, but also establishes the relationships between Pawlak’s rough sets and fuzzy sets or probabilistic rough sets. However, it is noteworthy that Pawlak’s rough membership functions have limitations when handling incomplete data that exist widely in the real world. As will be shown in this paper, one way to overcome this is to construct rough membership functions for covering-based rough sets. In this paper, we first use an example in evidence-based medicine to illustrate how to use Pawlak’s rough membership function on numerically characterizing decisions under the circumstances where data are complete. Then, we construct covering-based rough membership functions for four types of covering-based rough sets which were examined by Zhu and Wang (in IEEE Transactions on Knowledge and Data Engineering 19(8)(2007) 1131-1144 and Information Sciences 201(2012) 80-92), and use them to characterize these covering-based rough sets numerically. Finally, we present theoretical backgrounds for these covering-based rough membership functions, and illustrate how to apply them on numerically characterizing decisions under the circumstances where data are incomplete.

In this paper, we propose a new fuzzy forecasting method based on two-factors second-order fuzzy-trend logical relationship groups (TSFTLRGs), particle swarm optimization (PSO) techniques and similarity measures between the subscripts of fuzzy sets (FSs) for forecasting the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX) and the New Taiwan Dollar/US Dollar (NTD/USD) exchange rates. First, we propose a PSO-based optimal-intervals partition algorithm to get the optimal partition of the intervals in the universe of discourse (UOD) of the main factor TAIEX and to get the optimal partition of the intervals in the UOD of the secondary factor SF, where SF ź {Dow Jones, NASDAQ, M1B}. Based on the proposed PSO-based optimal-intervals partition algorithm, the constructed TSFTLRGs, and similarity measures between the subscripts of FSs, we propose a new method for forecasting the TAIEX and the NTD/USD exchange rates. The main contribution of this paper is that we propose a new fuzzy forecasting method based on TSFTLRGs, PSO techniques and similarity measures between the subscripts of FSs for forecasting the TAIEX and the NTD/USD exchange rates to get higher forecasting accuracy rates than the ones of the existing fuzzy forecasting methods.

Deep Neural Networks for image/video classification have obtained much success in various computer vision applications. Existing deep learning algorithms are widely used on RGB images or video data. Meanwhile, with the development of low-cost RGB-D sensors (such as Microsoft Kinect and Xtion Pro-Live), high-quality RGB-D data can be easily acquired and used to enhance computer vision algorithms [14]. It would be interesting to investigate how deep learning can be employed for extracting and fusing features from RGB-D data. In this paper, after briefly reviewing the basic concepts of RGB-D information and four prevalent deep learning models (i.e., Deep Belief Networks (DBNs), Stacked Denoising Auto-Encoders (SDAE), Convolutional Neural Networks (CNNs) and Long Short-Term Memory (LSTM) Neural Networks), we conduct extensive experiments on five popular RGB-D datasets including three image datasets and two video datasets. We then present a detailed analysis about the comparison between the learned feature representations from the four deep learning models. In addition, a few suggestions on how to adjust hyper-parameters for learning deep neural networks are made in this paper. According to the extensive experimental results, we believe that this evaluation will provide insights and a deeper understanding of different deep learning algorithms for RGB-D feature extraction and fusion.

Clustering analysis is a fundamental technique in machine learning, which is also widely used in information granulation. Multiple clustering systems granulate a data set into multiple granular structures. Therefore, clustering ensemble can serve as an important branch of multigranulation information fusion. Many approaches have been proposed to solve the clustering ensemble problem. This paper focuses on the direct approaches which involve two steps: finding cluster correspondence and utilizing a fusion strategy to produce a final result. The existing direct approaches mainly discuss the process of finding cluster correspondence, while the fusing process is simply done by voting. In this paper, we mainly focus on the fusing process and propose a Dempster-Shafer evidence theory-based clustering ensemble algorithm. The advantage of the algorithm is that the information of an object's surrounding cluster structure is taken into consideration by using its neighbors to describe it. First, we find neighbors of each object and generate its label probability outputs in every base partition. Second, these label probability outputs are integrated based on DS theory. Theoretically, our method is superior to other voting methods. Besides, several experiments show that the proposed algorithm is statistically better than seven other clustering ensemble methods.

In MMMF, ratings matrix with multiple discrete values is treated by specially extending hinge loss function to suit multiple levels.We view this process as analogous to extending two-class classifier to a unified multi-class classifier.Alternatively, multi-class classifier can be built by arranging multiple two- class classifiers in a hierarchical manner.In this paper, we investigate this aspect for collaborative filtering and propose an efficient and novel framework of multiple bi-level MMMFs.We compare our method with nine well-known algorithms on two benchmark datasets and show that our method outperforms these methods on NMAE measure. Maximum Margin Matrix Factorization (MMMF) has been a successful learning method in collaborative filtering research. For a partially observed ordinal rating matrix, the focus is on determining low-norm latent factor matrices U (of users) and V (of items) so as to simultaneously approximate the observed entries under some loss measure and predict the unobserved entries. When the rating matrix contains only two levels (1), rows of V can be viewed as points in k-dimensional space and rows of U as decision hyperplanes in this space separating +1 entries from 1 entries. When hinge/smooth hinge loss is the loss function, the hyperplanes act as maximum-margin separator. In MMMF, rating matrix with multiple discrete values is treated by specially extending hinge loss function to suit multiple levels. We view this process as analogous to extending two-class classifier to a unified multi-class classifier. Alternatively, multi-class classifier can be built by arranging multiple two-class classifiers in a hierarchical manner. In this paper, we investigate this aspect for collaborative filtering and propose an efficient and novel framework of multiple bi-level MMMFs. There is substantial saving in computational overhead. We compare our method with nine well-known algorithms on two benchmark datasets and show that our method outperforms these methods on NMAE measure. We also show that our method yields latent factors of lower ranks and the trade-off between empirical and generalization error is low.

We confirm that agent-ants can learn only via the pheromone trail.It is better to use both the heuristic function and the pheromone trail.The ACDT alg. makes it possible to optimize the construction of decision trees based on a heuristic.The pheromone map clearly present a data set as the solution space and pheromone trail values.Analysis of pheromone maps shows that the ACDT alg. searches different subspaces of the solution space. This paper considers the topic of cooperation between agent ants in an Ant Colony Optimization (ACO) algorithm that is used to construct decision trees (Ant Colony Decision Tree or ACDT). To follow a suitable methodology, the paper presents a formal definition of the ACDT algorithm with a focus on the influence that Ant Colony Optimization algorithms have on the obtained results. The aim of this paper is to provide the rationale for using swarm intelligence (i.e., ACO) in the process of constructing decision trees. Many experiments were conducted to provide a solid justification. These experiments tested cooperation between agent ants in ant colony algorithms with different ACO performance scenarios: the application of only a pheromone trail, the application of only a heuristic function, the application of both components, and the application of neither component. Additionally, different values of the pheromone trail were tested at various stages of the algorithm's operation and pheromone representations were presented. The experiments were conducted on 30 publicly available data sets; all observations were preceded by statistical tests. This paper confirms that it is reasonable to use the pheromone trail and balanced heuristics. Moreover, we found that, for the ACDT algorithm, good results can also be obtained without heuristics.

In this paper we propose a design for the kNN rule classifier based on GPU devices.This design overcomes the dependency between dataset size and GPU memory required.An efficient CPU-GPU communication scheme is designed.The carefully designed kernels for each step of the computation improve the run-time performance.An efficient selection algorithm replaces the sorting algorithm traditionally used to find the neighborhood. The k nearest neighbor (kNN) rule is one of the most used techniques in data mining and pattern recognition due to its simplicity and low identification error. However, the computational effort it requires is directly related to the dataset sizes, hence delivering a poor performance on large datasets. The use of graphics processing units (GPU) has improved the run-time performance of the kNN rule but the computational requirements of current approaches limit this performance as the dataset size increases.In this paper, we propose a new scalable and memory efficient design for a GPU-based kNN rule, called GPU-SME-kNN, that breaks the dependency between dataset size and memory footprint while delivering high performance. An experimental study of GPU-SME-kNN is presented showing a high performance, even in cases that other methods cannot address, while the computational requirements are suitable for most commercial GPU devices. Our design has also been applied to kNN-based lazy learning algorithms reducing run-times in a significant way.

Paper presents the work on development and validation of spatio-temporal clustering technique called ST-OPTICS.Emphasis has also been laid down on validation using performance indices.Capable to handle all types of dimensions i.e. spatial, non-spatial and temporal.It gives micro level clusters, agglomeration has been done for better visualization and interpretation.Results (average runtime & quality of clusters) found to be better than existing technique. Spatio-temporal data mining (STDM) is a process of the extraction of implicit knowledge, spatial and temporal relationships, or other patterns not explicitly stored in spatio-temporal databases. As data are growing not only from static view point, but they also evolve spatially and temporally which is dynamic in nature that is the reason why this field is now becoming very important field of research. In addition, spatio-temporal (ST) data tend to be highly auto-correlated, which leads to failure of assumption of independence, which is there in Gaussian distribution model. Vital issues in spatio temporal clustering technique for earth observation data is to obtain clusters of, good quality, arbitrary shape, problem of nested clustering and their validation. The present paper addresses these issues and proposes their solutions. In this direction, an attempt has been made to develop a clustering algorithm named as "Spatio-Temporal - Ordering Points to Identify Clustering Structure (ST-OPTICS)" which is modified version of existing density based technique - "Ordering Points to Identify Clustering Structure (OPTICS)". Experimental work carried out is analysed and found that quality of clusters obtained and run time efficiency are much better than existing technique i.e. ST-DBSCAN. In order to improve the visualization and the interpretation of obtained micro level clusters, sincere effort has been put in to merge the obtained clusters using agglomerative approach. Performance evaluation is done in both ways i.e. qualitatively and quantitatively for cross validating the results. Results show performance improvement of proposed ST-OPTICS clustering technique compared to ST-DBSCAN algorithm.

Fuzzy numbers are ranked using Dempster-Shafer theory (DST) with fuzzy targets.Fuzzy targets are used to reflect human viewpoints and avoid biases in fuzzy ranking.Fuzzy numbers are ranked with different viewpoints and with a combined viewpoint.Transferable Belief Model is adopted to elaborate DST.Benchmark examples are adopted for experimental studies and positive results are obtained. In this paper, an extended ranking method for fuzzy numbers, which is a synthesis of fuzzy targets and the Dempster-Shafer Theory (DST) of evidence, is devised. The use of fuzzy targets to reflect human viewpoints in fuzzy ranking is not new. However, different fuzzy targets can lead to contradictory fuzzy ranking results; making it difficult to reach a final decision. In this paper, the results from different viewpoints are treated as different sources of evidence, and Murphy's combination rule is used to aggregate the fuzzy ranking results. DST allows fuzzy numbers to be compared and ranked while preserving their uncertain and imprecise characteristics. In addition, a hybrid method consisting of fuzzy targets and DST with the Transferable Belief Model is formulated, which fulfils a number of important ordering properties. A series of empirical experiments with benchmark examples has been conducted and the experimental results clearly indicate the usefulness of the proposed method.

We evaluate network reliability where arc failure rates are fuzzy numbers.A network topology decomposition model based on cellular automata is proposed.The network arc failure rates following different types of membership function.Simulation based model is imprecise for lower probability estimations. This paper concentrates on the performance evaluation of networks, whose arc failure rates are not deterministic numbers, but imprecise ones. Conventional literatures analyze the network reliability assuming that the failure rates of all components in networks following the same membership function. However, most real-world networks do not abide by this regulation, especial complex ones. Therefore, in this paper, a new method is developed based on cellular automata (CA) and fuzzy logic following different types of fuzzy failure rates. The proposed method has two separate processing procedures: the computing of fuzzy numbers (CFN) and the decomposing model based on cellular automata (DB-CA). The combination of these procedures can be used to topology decomposition and the fuzzy reliability evaluation of any real network. Finally, The proposed method is improved by numerical network examples with a variety of membership functions, and the benchmark water delivery network in Shelby county is provided to estimate the feasibility and effectiveness of the proposed model.

Privacy and accountability are two important issues in mobile e-commerce. In this paper, we formalized the definition and security model of identity-based plaintext-checkable encryption (IBPCE) schemes, and proposed an IBPCE scheme. Furthermore, we designed an accountable mobile e-commerce (AMEC) scheme. This scheme can provide the following features: (1) users can register to the e-commerce system by using their mobile identities, such as mobile phone numbers; (2) the transactions between a buyer and a vendor are encrypted; (3) if there is a dispute, an offline adjudicator can identify who is dishonest by checking the encrypted transactions. Hence, our scheme is an ideal primitive to balance the need for privacy and accountability. In mobile e-commerce systems, users conduct transactions using wireless or Internet-based devices, such as mobile phones and tablets. It is different from traditional e-commerce systems relying on workstations or desktops, which is usually used in a fixed location. Recently, privacy and accountability have become users' primary concerns in mobile e-commerce applications. In this paper, a novel mobile e-commerce scheme is developed to address the fundamental requirements. We first propose an identity-based plaintext-checkable encryption (IBPCE) scheme where anyone can check whether a ciphertext is the encryption of a plaintext under a specific identity without knowing the secret key. Furthermore, the proposed IBPCE scheme is incorporated into the mobile e-commerce scenario, which results in an accountable mobile e-commerce (AMEC) scheme. Our proposed AMEC scheme has several superior features: (1) Users can register to the e-commerce system by using their mobile identities, such as mobile phone numbers; (2) The transactions between a buyer and a vendor are encrypted; (3) If there is a dispute, an offline adjudicator can identify who is dishonest by checking the encrypted transactions. We evaluate the proposed scheme and confirm that the new scheme can effectively balance the need for privacy and accountability.

Recommender systems apply machine learning techniques for filtering unseen information and can predict whether a user would like a given item. In this paper, we propose a new algorithm that we call the Kernel-Mapping Recommender (KMR), which uses a novel structure learning technique. This paper makes the following contributions: we show how (1) user-based and item-based versions of the KMR algorithm can be built; (2) user-based and item-based versions can be combined; (3) more information—features, genre, etc.—can be employed using kernels and how this affects the final results; and (4) to make reliable recommendations under sparse, cold-start, and long tail scenarios. By extensive experimental results on five different datasets, we show that the proposed algorithms outperform or give comparable results to other state-of-the-art algorithms.

The performance of Hidden Markov Models (HMMs) targeted for complex real-world applications are often degraded because they are designed a priori using limited training data and prior knowledge, and because the classification environment changes during operations. Incremental learning of new data sequences allows to adapt HMM parameters as new data becomes available, without having to retrain from the start on all accumulated training data. This paper presents a survey of techniques found in literature that are suitable for incremental learning of HMM parameters. These techniques are classified according to the objective function, optimization technique and target application, involving block-wise and symbol-wise learning of parameters. Convergence properties of these techniques are presented along with an analysis of time and memory complexity. In addition, the challenges faced when these techniques are applied to incremental learning is assessed for scenarios in which the new training data is limited and abundant. While the convergence rate and resource requirements are critical factors when incremental learning is performed through one pass over abundant stream of data, effective stopping criteria and management of validation sets are important when learning is performed through several iterations over limited data. In both cases managing the learning rate to integrate pre-existing knowledge and new data is crucial for maintaining a high level of performance. Finally, this paper underscores the need for empirical benchmarking studies among techniques presented in literature, and proposes several evaluation criteria based on non-parametric statistical testing to facilitate the selection of techniques given a particular application domain.

Mapping nodes in multi-layer and heterogeneous networks to low-dimensional vectors has wide applications in community detection, node classification and link prediction, etc. In this paper, a generalized graph representation learning framework is proposed for information aggregation in various multi-layer and heterogeneous networks. Specifically, an aggregation network is firstly obtained by graph transformation, generating potential information links based on the network structure on different layers. A comprehensive measurement of the similarity between different nodes in the aggregation network is then carried out by aggregating the information of nodes’ identities of structure, nearness and attributes etc. Based on the comprehensive similarity values the nodes have, a context graph can be generated using a simple edge percolation method, which provides a basis facilitating some important downstream work such as classification, clustering and prediction etc. We demonstrate the effectiveness of the new framework in identifying subnetworks in a cyberspace network, where it significantly outperforms all the existing baselines. Ministry of Education (MOE) G.Y. and Y.K. were supported by NSSFC 2019-SKJJ-C-005. G.X. was supported by the Ministry of Education (MOE), Singapore, under contract RG19/20.

Leaks in water distribution networks cause a loss of water that needs to be compensated to ensure a continuous supply for all customers. This compensation is achieved by increasing the flow of the network, which entails an undesirable economical expense as well as negative consequences for the environment. For these reasons, detecting and fixing leaks is a relevant task for water distribution companies. This paper proposes a water leak detection method based on a self-supervised classification of flow time series. The aim is to detect the leaks in the network, providing a low false positive rate. The proposed method is applied to two water distribution networks and compared to two other methods in the literature, obtaining the best balance between the number of false positives and detected leaks.

In this paper, we study the propagation characteristics of rumors in an institution and develop the rumor dissemination model inside the institution. Different from the traditional model, a person in the proposed model can make a judgement about the message and decide to propagate it or refute it, and most people inside the institution can refute the rumor spreaders and propagate the genuine messages to uninformed people when they have confirmed the message is a rumor. Then, we split all the people into two institutions (inside and outside). Since the rumors and genuine messages from the institution can have a non-negligible impact on people outside the institution, we put forward a new double-institution rumor propagation model , and the model considers the impact of messages on the inside and outside of the institution simultaneously. Based on the two proposed models, the basic reproduction numbers are obtained respectively, and the local and global stability of the rumor-free equilibrium points are discussed separately. We numerically simulate the propagation of rumors in small-world networks. The simulation is carried out to verify the validity of the proposed model, and our model is closer to the reality than traditional models.

The Internet of Things (IoT) is commonly deployed with devices of limited power and computation capability. A centralized IoT architecture provides a simplified management for IoT system but brings redundancy by the unnecessary data traffic with a data center. A decentralized IoT reduces the cost on data traffic and is resilient to the single-point-of-failure. The blockchain technique has attracted a large amount of research, which is redeemed as a perspective of decentralized IoT system infrastructure. It also brings new privacy challenges for that the blockchain is a public ledger of all digital events executed and shared among all participants. The decentralized IoT system relies on the leader election deeply to implement the decentralized communications among the distributed nodes. The conventional leader election must have a centralized authority, contrasting to the decentralization. As an alternative, self-tallying type schemes have been proposed in the literature for decentralized systems. These schemes suffer from adaptive and abortive issues. Also, some additional factors should be considered, such as the availability of candidate nodes. If the candidate node is unavailable after the voting phase due to being offline or ongoing tasks, the next available candidate should be elected. To accommodate such a need, in this paper, we propose a new leader election paradigm called voluntary-tallying leader election, which achieves the core requirements such as ballet secrecy, voter privacy and the additional feature of voluntary-tallying. We formalize the system and security models for this new election paradigm and present a secure and practical construction.

Nowadays, crowdsourcing is a widespread and effective method to gather the crowd wisdom. At the same time, label aggregation is used to aggregate the noisy and biased data generated by the crowd. In the real-world crowdsourcing tasks, most workers only answer a small fraction of questions, which makes the collected answer sparse. However, the existing label aggregation approaches often build upon some probabilistic modeling procedures which is sensitive to the data sparsity. In this paper, we exploit the predicted answers to improve the performance of label aggregation and propose PLA (Prediction-based Label Aggregation) to intelligently aggregate the crowd wisdom. With PLA, we firstly learn representations to capture the characteristics of the workers and questions. Then we deploy a neural network model to predict the answer given by different workers. After that we add the most valuable predicted answers to the answer set. Finally, we use the augmented answer set to enhance representative label aggregation algorithms. To validate our proposed PLA, we compare it with other 6 existing methods on 8 real-world datasets. Our results show that PLA can enhance the performance of different aggregation algorithms in crowdsourcing tasks and achieves up to 16% performance improvement.

The data clustering problem can be described as the task of organizing data into groups, where in each group the objects share some similar attributes. Most of the problems clustering algorithms address do not have a prior solution. This paper addresses the algorithm selection challenge for data clustering, while taking the difficulty in evaluating clustering solutions into account. We present a new meta-learning method for recommending the most suitable clustering algorithm for a dataset. Based on concepts from the isolation forest algorithm, we propose a new similarity measure between datasets. Our proposed dataset characterization methods generate an embedding for a dataset using this similarity measure, which is then used to improve the quality of the problem’s characterization. The method utilizes landmarking concepts to characterize the dataset and then, inspired by the DeepFM algorithm, applies meta-learning to rank the candidate algorithms that are expected to perform the best for the current dataset. This ranking could, among other things, support AutoML systems. Our approach is evaluated on a corpus of 100 publicly available benchmark datasets. We compare our method’s ranking performance to that of existing meta-learning methods and show the dominance of our method in terms of predictive performance and computational complexity.

In many real-world applications, it is the fairness, not the accuracy, of a machine learning (ML) classifier that is the crucial factor. Post-processing approaches are widely considered as successful tools to improve the fairness of black-box ML classifiers. These aim to learn a relabeling function to modify initial predicted labels provided by a pre-trained “unfair” classifier, resulting in fair classification on a given test set. However, many post-processing methods require a training set with true labels to learn the relabeling function. To the best of our knowledge, there have been only two methods that learn the relabeling function without requiring the true labels of training samples. However, both of these methods require access to the predictions of the pre-trained classifier when performing on the test set, even after they learned the optimal relabeling function, and neither offers theoretical guarantees on the trade-off between accuracy loss and fairness improvement. In this paper, we propose a novel post-processing method based on Gaussian process (GP). We first train a GP with unlabeled samples, and use its posterior mean function to approximate the predictions of the pre-trained classifier. We then adjust the mean function (i.e. the relabeling function) to achieve two goals: (1) maximize the fairness and (2) minimize the difference between the relabeling function and the pre-trained classifier. By doing this, our method can improve fairness while maintaining high accuracy. We provide a theoretical analysis to derive an upper bound on accuracy loss for our method. We demonstrate our method on four real-world datasets, comparing with state-of-the-art baselines, to demonstrate its ability to achieve both fairness and accuracy.

This paper presents an efficient graph semisupervised learning (GSSL) method that meets the criterion of optimization without iterations. Most existing GSSL methods require iterative optimization to achieve a preset objective because they consider data points to be in peer-to-peer relationships. Additionally, existing GSSL methods must learn from scratch for unseen data because graph structures are specifically built for a given dataset. By leveraging the partial order relationships induced by the local density and distances between data, we developed a novel label propagation algorithm based on the data structure of an optimal leading forest (OLeaF). The time complexity of our method is O(N) for both labeling unclassified data and labeling new data from a dataset after an OLeaF is constructed. Therefore, the two main weaknesses of traditional GSSL are addressed. Additionally, the constructed leading forest offers good interpretability for learning results. We scale the proposed method to accommodate big data by utilizing the block distance matrix technique and locality-sensitive hashing. Extensive experiments on datasets with different characteristics demonstrate the superior efficiency and competitive accuracy of the proposed method.

Unsupervised attribute reduction becomes very challenging due to a lack of decision information, which is to select a subset of attributes that can maintain learning ability without decision information. However, most of the existing unsupervised attribute reduction methods are proposed for numerical or nominal attributes, and little research has been done on unsupervised mixed attribute reduction methods. In view of this, this paper proposes a generalized unsupervised mixed attribute reduction model based on fuzzy rough sets . First, based on all single attribute subsets , the significance is defined to indicate the importance of a candidate attribute. Then, a specific fuzzy rough-based unsupervised attribute reduction (FRUAR) algorithm is designed. Finally, the proposed algorithm is compared with the existing algorithms by using thirty public data sets. Experimental results show that the algorithm FRUAR can select fewer attributes to maintain or improve the performance of learning algorithms, and it is suitable for mixed attribute data.

Network embedding has been successfully used for a variety of tasks, e.g., node clustering, community detection, link prediction and evolution analysis on complex networks. For a given network, embedding methods are usually designed based on first-order proximity, second-order proximity, community constraints, etc. However, they are incapable of capturing the structural similarity of nodes. The bridge nodes with small proximity and located in different communities, should be similar in embedding space since they have the same surrounding structure. In this paper, these structural features are referred to as lower-order information, which could reveal and modify the structural similarity of nodes in the embedding space. Specifically, we propose to construct the feature matrix with the lower-order information of the network. In order to effectively fuse the structural features of nodes into embedding space, an intuitive, interpretable and feasible method named LONE-NMF is proposed, which adopts the representation learning framework based on non-negative matrix factorization. It can effectively learn the representation vectors of nodes in the network via preserving the proximity and lower-order information. Moreover, an optimization algorithm is designed for LONE-NMF. Extensive experiments based on clustering and link prediction show that the proposed method achieves significant performance improvement comparing with some baselines. Finally, we validate the principle and advantage of LONE-NMF through a case study.