Search

In this paper, a sentiment classification model is proposed to address two predominant issues in sentiment classification, namely domain-sensitive and data imbalance. Since words may embed distinct sentiment polarities in different contexts, sentiment classification is widely contended as a domain-sensitive task. Accordingly, this paper draws on label propagation to induce universal and domain-specific sentiment lexicons and builds a domain-adaptive sentiment classification model that incorporates universal and domain-specific knowledge into a unified learning framework. On the flip side, sentiment-related corpuses are usually formed with skewed polarity distribution because individuals tend to share similar assessment criteria on a given object and hence their sentiment polarities toward the same object are likely to be similar. We endeavor to address such imbalanced data problem by advancing a novel over-sampling technique. Unlike existing over-sampling approaches that generate minority-class samples from numerical feature space, the proposed sampling method directly creates synthetic texts from word spaces. Several experiments are conducted to verify the effectiveness of the proposed lexicon generation method, learning framework, and over-sampling method. Results show that the induced sentiment lexicons are interpretable and the proposed model is found to be effective for imbalanced and domain-specific text sentiment classification.

This paper addresses the problem of a finite set of entities which are required to achieve a goal within a predefined deadline. For example, a group of students is supposed to submit a homework by a specified cutoff. Further, we are interested in predicting which entities will achieve the goal within the deadline. The predictive models are built based only on the data from that population. The predictions are computed at various time instants by taking into account updated data about the entities. The first contribution of the paper is a formal description of the problem. The important characteristic of the proposed method for model building is the use of the properties of entities that have already achieved the goal. We call such an approach “Self-Learning”. Since typically only a few entities have achieved the goal at the beginning and their number gradually grows, the problem is inherently imbalanced. To mitigate the curse of imbalance, we improved the Self-Learning method by tackling information loss and by several sampling techniques. The original Self-Learning and the modifications have been evaluated in a case study for predicting submission of the first assessment in distance higher education courses. The results show that the proposed improvements outperform the specified two base-line models and the original Self-Learner, and also that the best results are achieved if domain-driven techniques are utilised to tackle the imbalance problem. We also showed that these improvements are statistically significant using Wilcoxon signed rank test.

Decision-making is one of the important deals in knowledge transfer and it can be useful for the multi-source domains. Meanwhile, the existing knowledge transfer schemes do not use privacy-preserving techniques for preserving security. This can be a problem for critical domains like financial market forecasting as the misuse of security can lead to legal and financial implications. In recent years, cloud services have revolutionized various technological applications. Cloud computing has become more popular with digital technologies as it provides uninterrupted services like transmission, storage, and intensive computing of data. The architecture of the cloud is also cost-efficient. Besides, various promising services from the cloud, some challenges need to be addressed to secure the privacy of the cloud users as millions of users access its services. Privacy preservation is an important aspect in the field of data mining, and the necessity of securing important data in the cloud from hackers is on the rise. Privacy-preserving data mining algorithms have been analyzed over recent years to provide sufficient solutions for securing the privacy of the data in the cloud. This paper plans to introduce a new hybrid meta-heuristic concept for developing a privacy preservation strategy towards business data under the cloud sector. The main objective of this paper is to design a new hybrid red deer-bird swarm algorithm (RD-BSA) to ensure higher convergence and since the use of control parameters over the solution generation is minimized. The proposed privacy preservation scheme on three financial databases is evaluated with the performance against the existing privacy preservation schemes. Different analyses like statistical, key sensitivity, Known-Plaintext Attack (KPA), and Chosen-Plaintext Attack (CPA) are used for evaluating the efficiency of the algorithm. The comparative analysis of the proposed model over the conventional models demonstrates its effective performance via diverse analysis.

Recommender Systems (RS) provide a relevant tool to mitigate the information overload problem. A large number of researchers have published hundreds of papers to improve different RS features. It is advisable to use RS frameworks that simplify RS researchers: a) to design and implement recommendations methods and, b) to speed up the execution time of the experiments. In this paper, we present CF4J, a Java library designed to carry out Collaborative Filtering based RS research experiments. CF4J has been designed from researchers to researchers. It allows: a) RS datasets reading, b) full and easy access to data and intermediate or final results, c) to extend their main functionalities, d) to concurrently execute the implemented methods, and e) to provide a thorough evaluation for the implementations by quality measures. In summary, CF4J serves as a library specifically designed for the research trial and error process.

With the massive explosion of machine learning in our day-to-day life, incremental and adaptive learning has become a major topic, crucial to keep up-to-date and improve classification models and their corresponding feature extraction processes. This paper presents a categorized overview of incremental feature extraction based on linear subspace methods which aim at incorporating new information to the already acquired knowledge without accessing previous data. Specifically, this paper focuses on those linear dimensionality reduction methods with orthogonal matrix constraints based on global loss function, due to the extensive use of their batch approaches versus other linear alternatives. Thus, we cover the approaches derived from Principal Components Analysis, Linear Discriminative Analysis and Discriminative Common Vector methods. For each basic method, its incremental approaches are differentiated according to the subspace model and matrix decomposition involved in the updating process. Besides this categorization, several updating strategies are distinguished according to the amount of data used to update and to the fact of considering a static or dynamic number of classes. Moreover, the specific role of the size/dimension ratio in each method is considered. Finally, computational complexity, experimental setup and the accuracy rates according to published results are compiled and analyzed, and an empirical evaluation is done to compare the best approach of each kind.

Social sensing has emerged as a new data collection paradigm in networked sensing applications where humans are used as “sensors” to report their observations about the physical world. While many previous studies in social sensing focus on the problem of ascertaining the reliability of data sources and the correctness of their reported claims (often known as truth discovery), this paper investigates a new problem of critical source selection. The goal of this problem is to identify a subset of critical sources that can help effectively reduce the computational complexity of the original truth discovery problem and improve the accuracy of the analysis results. In this paper, we propose a new scheme, Critical Source Selection (CSS), to find the critical set of sources by explicitly exploring both dependency and speak rate of sources. We evaluated the performance of our scheme and compared it to the state-of-the-art baselines using two data traces collected from a real world social sensing application. The results showed that our scheme significantly outperforms the baselines by finding more truthful information at a higher speed.

Traditionally, research on network theory focused on studying graphs with equivalent entities failing to deliberate the useful supplementary information related to the dynamic properties of the complex network interactions. This paper tries to study the evolution process of dynamic complex networks from a multilayer perspective by analyzing the properties of naturally multilayered web-based directed complex social networks of Google+ and Twitter, and undirected collaborative networks of DBLP and ASTRO-PH, thereby proposing a new non-parametric knowledge-based multilayer link recommendation approach. The paper investigates the layers’ evolution throughout the network evolution, inspects the evolution of each node's membership in different layers by an Infinite Factorial Hidden Markov Model, and finally formulates the intra-layer and inter-layer link generation process. Some Markov Chain Monte Carlo sampling strategies are driven to simulate parameters of the proposed multilayer model, using certain synthetic and real complex network datasets. Experimental results indicate great improvements in the performance of the proposed multilayer link recommendation approach in terms of certain analyzed performance measures.

Multiple attribute decision making (MADM) problems often include quantitative and qualitative attributes which can be assessed by numerical values and subjective judgements respectively. The evidential reasoning (ER) rule provides a process for dealing with this type of MADM problems of both a quantitative and qualitative nature of uncertainty. In this paper, the ER rule is generalized to dealing with MADM problems in group decision making circumstance where the weights and reliabilities of both experts and attributes are considered. Specifically, the result and process aggregation based ER rules for MADM in group decision making are given respectively, followed by the comparative analysis on the given aggregations. The ER analytical rule for group MADM problems is also provided for the generalization of the ER analytical approach where group decision making is not considered. It is also a development of Yang's ER rule which is a recursive calculation process. Due to the fact that uncertainty and ambiguity are always existent in group decision making, interval weights and reliabilities of experts and attributes should be taken into account in the process of experts’ judgment aggregation. In this paper, several ER based programming models under interval weights and reliabilities are constructed for the generation of global belief degrees in a consistent way. A case study is conducted on the life cycle assessment of electric vehicles to illustrate the applicability of the proposed method and the potential in supporting MADM in group decision making.

This paper focuses on concept drift in business intelligence and data-driven decision support systems (DSSs). The assumption of a fixed distribution in the data renders conventional static DSSs inaccurate and unable to make correct decisions when concept drift occurs. However, it is important to know when, how, and where concept drift occurs so a DSS can adjust its decision processing knowledge to adapt to an ever-changing environment at the appropriate time. This paper presents a data distribution-based concept drift detection method called fuzzy competence model drift detection (FCM-DD). By introducing fuzzy sets theory and replacing crisp boundaries with fuzzy ones, we have improved the competence model to provide a better, more refined empirical distribution of the data stream. FCM-DD requires no prior knowledge of the underlying distribution and provides statistical guarantee of the reliability of the detected drift, based on the theory of bootstrapping. A series of experiments show that our proposed FCM-DD method can detect drift more accurately, has good sensitivity, and is robust.

Knowledge is the foundation of emergency decision-making (EDM), in which the experts from multi-fields express their knowledge with multi-granularity linguistic model to assistant decision-making. Thus, the paper proposed a new decision support methodology to generate decision-making knowledge. In this paper, the framework of decision knowledge generation in the EDM was introduced firstly. To generate decision-making knowledge accurately and objectively, two objective models, which can effectively determine the weights of criteria and experts respectively, were built based on the tacit knowledge hidden in the original information. Then, the personal knowledge, generated by combining the normalized decision knowledge and the weight vector of criteria, is further aggregated into the collective knowledge by means of aggregation operator. Finally, an illustrative example is presented to verify the application of the proposed methods, and relevant discussions prove the results obtained from the proposed decision support methodology can improve the scientificity and accuracy of the EDM.

Multiple instance learning (MIL) is a generalization of supervised learning which attempts to learn a distinctive classifier from bags of instances. This paper addresses the problem of the transfer learning-based multiple instance method for text categorization problem. To provide a safe transfer of knowledge from a source task to a target task, this paper proposes a new approach, called selective multiple instance transfer learning (SMITL), which selects the case that the multiple instance transfer learning will work in step one, and then builds a multiple instance transfer learning classifier in step two. Specifically, in the first step, we measure whether the source task and the target task are related or not by investigating the similarity of the positive features of both tasks. In the second step, we construct a transfer learning-based multiple instance method to transfer knowledge from a source task to a target task if both tasks are found to be related in the first step. Our proposed approach explicitly addresses the problem of safe transfer of knowledge for multiple instance learning on the text classification problem. Extensive experiments have shown that SMITL can determine whether the two tasks are related for most data sets, and outperforms classic multiple instance learning methods.

In this paper, an ontology-supported case-based reasoning approach for computer-aided tolerance specification is proposed. This approach firstly considers the past tolerance specification problems and their schemes as previous cases and the new tolerance specification problems as target cases and uses an ontology to represent previous and target cases. Then certain ontology-based similarity measure is used to assess the similarity between the toleranced features of target and previous cases, the similarity between the part features of target and previous cases, and the similarity between the topological relations of target and previous cases. Based on these similarities, an ontology-based similarity measure for computing the similarity between target and previous cases is designed, and an algorithm for establishing such similarity measure with high accuracy and retrieving similar previous cases for a target case with this similarity measure is presented. This algorithm shows how to linearly combine the similarity of toleranced features, the similarity of part features, and the similarity of topological relations to assess the similarity between target and previous cases to implement retrieval of previous cases under the prerequisite of ensuring the highest accuracy of the similarity measure. The paper also reports a prototype implementation of the proposed approach, provides an example to illustrate how the approach works, and evaluates the approach via theoretical and experimental comparisons.

The three-way decision model is a topic of substantial research interest in the field of artificial intelligence, and many researchers have focused on to its feasibility and rationality. The tolerance and practicability of the three-way decision model are better than those of the two-way decision model. When the attribute value of each object in a domain is given, the formation of a three-way classification of the domain is a key issue. However, few studies have been conducted on establishing a three-way decision model with the given attribute values in the case where the number of objects in an accepted region is given. Therefore, in the model presented in this paper, both the uncertainty of attribute values and the cost of updating are fully considered. In this paper, first, a new concept of attribute ratio is defined to describe an object when the attribute value of the object is numerical, and then, a dynamic three-way decision model is established. Second, a feature extraction algorithm of attribute values is proposed, and a pair of decision thresholds of the dynamic three-way decision model is also obtained according to the given conditions. Then, in the case where the attribute values are updated, an example is provided to demonstrate how two-way classification results can be obtained in the dynamic decision-making process. Finally, the results of simulation experiments show that the proposed model is feasible and effective in practical applications. When the number of objects in an accepted region has been given, according to the updating strategy of attribute values, the three-way decision problems are successfully solved by the proposed model.

The Extended Belief Rule-Based (EBRB) system has been widely used to solve the real-world problems concerning with incompleteness, uncertainty, and ambiguity. However, EBRB is essentially a data-driven method, in which each rule is obtained from training data. Therefore, the generated extended belief rules may be severely biased when dealing with data with imbalanced classes. In this case, the number of the rules generated by the samples of majority classes (i.e., negative samples) may be much larger than those of minority classes (i.e., positive samples). Thus, the class imbalance may lead to significant biases in system decision-making. In order to resolve this problem, this paper proposes a novel EBRB system based on fuzzy C-means clustering (FCM-EBRB). First, we adopt FCM clustering to oversample the positive samples and undersample the negative ones, so as to achieve the balance between them. Next, this paper improves the construction method of EBRB and optimizes the system through an efficient parameter learning strategy. Finally, this paper conducts comprehensive comparison experiments on a binary classification synthetic dataset and 11 commonly used KEEL public class imbalance datasets. Experimental results show that the proposed method can effectively reduce the scale of the rule base and achieve high inference accuracy, especially for imbalanced data.

Universum support vector machine ( U -SVM), due to its tremendous accuracy improvements, has been expanded and applied in all kinds of fields. Universum encodes related prior knowledge but does not belong to any class of interest. With Universum, the number of training samples and computational complexity are clearly increased. Inspired by the sparsity of SVMs, a safe sample screening rule (SSSR) for U -SVM is proposed in this paper. Our SSSR eliminates not only the labelled samples but also the Universum samples before training process, then the computational cost is dramatically reduced. Moreover, the same solution as the original problem can be obtained by utilizing our SSSR, that is, the training process is guaranteed to be accelerated safely. Besides, we extend our rule to the Universum twin support vector machine ( U -TSVM), and the SSSR for U -TSVM is also discussed in this paper. To the best of our knowledge, SSSR is the only existing safe screening method for U -SVMs. Numerical experiments on seventeen benchmark datasets, ABCDETC dataset and Chinese wine dataset demonstrate that the computational cost can be dramatically reduced without sacrificing the optimality of the final solution by our SSSR.

In this paper, we propose a deep collaborative filtering recommendation model, which consists of an attention-based representation learning component and a multi-input matching function learning component. This model takes interaction matrix based on implicit feedback as data source to construct representations of long-term user preferences and item latent features. In the representation learning, a time-aware attention network is used, which uses the embedding vectors of the predicted item, recent historical interaction items, and the interaction time of recent historical interaction items to estimate the weights of different historical interaction items to short-term user preferences modeling. Then, the dynamic user preference representation can be obtained by combining short-term preferences with long-term preferences. In the matching function learning, a multi-input deep learning model is used. Its input includes not only the dynamic user preference representation and the item latent feature representation, but also the linear interaction between the two representations, so that the model has more powerful feature interactions learning ability. Experimental results on four datasets from different domains show that our method is largely superior to the state-of-the-art collaborative filtering methods, and the novel techniques we propose in this paper are effective in improving recommendation performance.

Cardiovascular disease (CVD) is a general term for a series of heart or blood vessels abnormality that serves as a global leading reason for death. The earlier the abnormal heart rhythm is discovered, the less severe the sequela and the faster the recovery. Electrocardiogram (ECG), as a main way to detect the electrical activity of heart, is a very important harmless means of predicting and diagnosing CVDs. However, ECG signal has characteristics of complex and high chaos, making it time-consuming and exhausting to interpret ECG signal even for experts. Hence, computer-aided methods are required to relief human burden and reduce errors caused by tiredness, inter- and intra-difference. Deep learning shows outstanding performance on ECG classification studies recent few years. Its hierarchical architecture enables higher-level features obtained and its strong ability to feature extraction contributes to classification project. Latest studies can achieve higher accuracy and efficiency than manual classification by experts. In this paper, we review the existing studies of deep learning applied in ECG diagnosis according to four typical algorithms: stacked auto-encoders, deep belief network, convolutional neural network and recurrent neural network. We first introduced the mechanism, development and application of the algorithms. Then we review their applications in ECG diagnosis systematically, discussing their highlights and limitations. Our view about future potential development of deep learning in ECG diagnosis is stated in the final part of this paper.

Multi-label text classification is an important yet challenging task in natural language processing. It is more complex than single-label text classification in that the labels tend to be correlated. To capture this complex correlations, sequence to sequence model has been widely applied, and achieved impressing performance for multi-label text classification. It encodes each document as contextual representations, and then decodes them to generate labels one by one. At each time step, the decoder usually adopts the attention mechanism to highlight important contextual representations to predict a related label, which has been proved to be effective. Nevertheless, the traditional attention approaches only utilize a hidden state to explore such contextual representations, which may result in prediction errors, or omit several trivial labels. To tackle this problem, in this paper, we propose “history-based attention”, which takes history information into consideration, to effectively explore informative representations for labels’ predictions in multi-label text classification. Our approach consists of two parts: history-based context attention and history-based label attention. History-based context attention considers historical weight trends to highlight important context words, which is helpful to predict trivial labels. History-based label attention explores historical labels to alleviate the error propagation problem. We conduct experiments on two popular text datasets (i.e., Arxiv Academic Paper Dataset and Reuters Corpus Volume I), it is demonstrated that the history-based attention mechanism could boost the performance to a certain extent, and the proposed method consistently outperforms highly competitive approaches.

With the vigorous development of the information technology industry, the information data available to mankind has shown an explosive growth trend. Dynamic concept learning is an approach that can effectively process the acquired massive data and extract valuable information from them. Concept cognitive learning (CCL) is a very active research direction in the field of dynamic concept learning, while partial order formal structure analysis (POFSA) is a concrete and practical model of CCL. However, the existing CCL algorithms in POFSA face some challenges when processing constantly changing data. Therefore, this paper is devoted to explore an incremental CCL algorithm based on three-way object partial order structure diagram (OPOSD) in POFSA with the incorporation of the thoughts of incremental learning. The features of five object categories are considered, and their incremental influences on three-way OPOSD are analyzed and their incremental CCL algorithms in three-way OPOSD are established. Based on some real famous formal contexts, this paper conducts numerical experiments, and the results show that the incremental CCL algorithm based on three-way OPOSD is consistent with human cognitive principles, and can improve the CCL performance of POFSA as well.

Along with the rapid development of network technologies, the server-side of an e-commerce website is becoming more and more untrustworthy. Thus, how to prevent the disclosure of users’ behavior privacy in online business activities has attracted people’s wide attention. Aiming at the protection of users’ commodity viewing privacy in a commercial website, this paper proposes to construct a group of dummy requests on a trusted client, then, which are submitted together with a user commodity viewing request to the untrusted server-side, so as to confuse and cover up the user preferences. First, we define a privacy model for a user commodity viewing service, in which we introduce a concept called entropy for commodity viewing probability to measure the confusion effect of dummy requests on user requests, and we introduce a concept called regional distance among commodity categories to measure the cover-up effect of dummy requests on users’ commodity viewing preferences. Second, we design an implementation algorithm to generate a group of ideal dummy requests that can meet the constraints formulated in the privacy model. Finally, both theoretical analysis and experimental evaluation demonstrate the effectiveness of the proposed approach, i.e., which can improve the security of users’ commodity viewing privacy on the untrusted server-side, without compromising the availability of an e-commerce website. In this paper, we present a valuable research attempt to the protection of users’ behavior privacy in a commercial website, which is of positive significance for building a privacy-preserving e-commerce platform.

This paper proposes a novel algorithm for the end-to-end recognition of slab identification numbers (SINs). In the steel industry, automatic recognition of an individual product information is important for production management. The recognition of SINs in actual factory scenes is a challenging problem due to complicated background and low-quality of characters. Conventional rule-based algorithms were developed to extract information of SINs, but these methods require engineering knowledge and tedious work for parameter tuning. The proposed algorithm employs a data-driven method to overcome these limitations and to handle the challenges for the recognition of SINs. This paper proposes accumulated response map and model-based score function to effectively use the outputs of a deep convolutional neural network. Experiments were thoroughly conducted for industrial data collected from an actual steelworks to verify the effectiveness of the proposed algorithm. Experiment results demonstrate that simultaneous recognition of entire characters in a SIN by optimizing the model-based score function is more effective for the robust performance compared to separated recognition of individual characters.

Hierarchically structured data are very common or even unavoidable for data mining and knowledge discovering from the perspective of granular computing in real-life world. Based on this circumstance, multi-scale information system is introduced by Wu and Leung and extends the theory and application of information system. In such table, objects may take different values under the same attribute measured at different scales. Recently, scale selection is the main issue of multi-scale information system, and optimal scale selection is to choose a proper decision table for final decision making or classification. In this paper, we firstly propose the concept of multi-scale attribute significance, and, in the sense of binary classification, another two equivalent definitions are given. Then based on the concept of significance, this paper introduces a novel approach of stepwise optimal scale selection to obtain one optimal scale combination with less time cost compared with the lattice model. Specially, for inconsistent multi-scale decision tables, different types of consistence are considered with different requirements for optimal scale selection. Finally, five algorithms are designed and six numerical experiments are employed to illustrate the feasibility and efficiency of the proposed model.

This paper demonstrates an innovative data analysis for weather using Cloud Computing, integrating both system and application Data Science services to investigate extreme weather events. Identifying five existing projects with ongoing challenges, our aim is to process, analyze and visualize collected data, study implications and report meaningful findings. We demonstrate the use of Cloud Computing technologies, MapReduce and optimization techniques to simulate temperature distributions and analyze weather data. Two major cases are presented. The first case is focused on forecasting temperatures based on studying trends from the historical data of Sydney, Singapore and London to compare the historical and forecasted temperatures. The second case is to use five-step MapReduce for numerical data analysis and eight-step process for visualization, which is used to analyze and visualize temperature distributions in the United States, before, during and after the time of experiencing polar vortex, as well as in the United Kingdom during and after the flood. Optimization was used in experiments involved up to 100 nodes between Cloud and non-Cloud and compared performance with and without optimization. There was an improvement in performance between 20% and 30% under 60 nodes in Cloud. Results, discussion and comparison were presented. We justify our research contributions and explain thoroughly in the paper how the three goals can be met: (1) forecasting temperatures of three cities based on evaluating the trends from the historical data; (2) using five-step MapReduce to achieve shorter execution time on Cloud and (3) using eight-step MapReduce with optimization to achieve data visualization for temperature distributions on US and UK maps.

Stochastic graph as a graph model for complex social networks.Four sampling algorithms for stochastic graphs in which edge weights are random variables.Analyze complex networks using stochastic network measures and sampling algorithms.Study the performance of the sampling algorithms on the stochastic networks. Recently, there has been growing interest in social network analysis. Graph models for social network analysis are usually assumed to be a deterministic graph with fixed weights for its edges or nodes. As activities of users in online social networks are changed with time, however, this assumption is too restrictive because of uncertainty, unpredictability and the time-varying nature of such real networks. The existing network measures and network sampling algorithms for complex social networks are designed basically for deterministic binary graphs with fixed weights. This results in loss of much of the information about the behavior of the network contained in its time-varying edge weights of network, such that is not an appropriate measure or sample for unveiling the important natural properties of the original network embedded in the varying edge weights. In this paper, we suggest that using stochastic graphs, in which weights associated with the edges are random variables, can be a suitable model for complex social network. Once the network model is chosen to be stochastic graphs, every aspect of the network such as path, clique, spanning tree, network measures and sampling algorithms should be treated stochastically. In particular, the network measures should be reformulated and new network sampling algorithms must be designed to reflect the stochastic nature of the network. In this paper, we first define some network measures for stochastic graphs, and then we propose four sampling algorithms based on learning automata for stochastic graphs. In order to study the performance of the proposed sampling algorithms, several experiments are conducted on real and synthetic stochastic graphs. The performances of these algorithms are studied in terms of Kolmogorov-Smirnov D statistics, relative error, Kendall's rank correlation coefficient and relative cost.

In this paper, we focus on the problem of context-aware recommendation using tensor factorization. Traditional tensor-based models in context-aware recommendation scenario only consider user-item-context interactions. In this paper, we argue that rating can't be totally explained by the interactions and the rating also influenced by the combined impact of overall mean, user bias, item bias and context bias. Based on this hypothesis, we propose a novel context-aware recommendation model named Bias Tensor Factorization, which take all this factors into account. Additionally, traditional context-aware recommenders with tensor factorization still have three main drawbacks: (1) the model complexity of those models increase exponentially with the number of context features, (2) those models can only handle context features with categorical values and (3) the models fail to select effective features from available context features. To address those problems, we propose a context features auto-encoding algorithm based on regression tree which can both handle numerical features and select effective features. Then we integrate this algorithm with Bias Tensor Factorization. Experiments on a real world contextual dataset and Movielens show that our proposed algorithms outperform the state-of-art context-aware recommendation algorithms, namely tensor factorization and factorization machine.

This paper proposes a new hybrid-coded HLO (HcHLO) framework to tackle mix-coded problems more efficiently and effectively.A new continuous human learning optimization algorithm is presented based on the linear learning mechanism of humans.The results show that the HcHLO achieves the best-known overall performance so far on the tested mix-coded problems. Human Learning Optimization (HLO) is an emerging meta-heuristic with promising potential, which is inspired by human learning mechanisms. Although binary algorithms like HLO can be directly applied to mixed-variable problems that contains both continuous values and discrete or Boolean values, the search efficiency and the performance of those algorithms may be significantly spoiled due to the curse of dimensionality caused by the binary coding strategy especially when the continuous parameters of problems require high accuracy. Therefore, this paper extends HLO and proposes a novel hybrid-coded HLO (HcHLO) framework to tackle mix-coded problems more efficiently and effectively, in which real-coded parameters are optimized by a new continuous HLO (CHLO) based on the linear learning mechanism of humans and the other variables are handled by the binary learning operators of HLO. Finally, HcHLO is adopted to solve 14 benchmark problems and its performance is compared with that of recent meta-heuristic algorithms. The experimental results show that the proposed HcHLO achieves the best-known overall performance so far on the test problems, which demonstrates the validity and superiority of HcHLO.

This paper proposed a new framework of one-class classification based on continuous cognition.The framework is implemented with LSSVM and the corresponding classifier is called OC-LSSVM.Several simulation and real datasets are used to test the performance of OC-LSSVM.OC-LSSVM shows state-of-the-art performance compared to established methods. One-class classification is a basic problem in machine learning. Unlike the existing typical one-class classifiers designed from the angle of probability or geometric, this paper attempts to study this problem from the bionics point of view. Using the continuous cognition characteristic as the starting point, we propose a new framework of one-class classifier, named multiple regression model (OC-MR), which can be seen as a natural extension of multiple regression for one-class classification problem. This paper applies least squares support vector machine (LSSVM) as an example to show themodeling process of the proposed method and the corresponding one-class classifier is named one-class least squares support vector machine (OC-LSSVM). Various simulation and real-life datasets are used to test the performance of the proposed OC-LSSVM. The existing popular one-class classification methods including Parzen kernel density estimation, support vector data description and Gaussian mixture model are also applied in order to achieve a comprehensive comparison. The results show that OC-LSSVM has achieved the best performance in most of the simulation and real-life datasets due to its good robustness, which highlights the efficacy of OC-LSSVM.

This paper develops a principled approach to accurately identify interesting places in a city through social sensing applications. Social sensing has emerged as a new application paradigm, where a crowd of social sources (humans or devices on their behalf) collectively contribute a large amount of observations about the physical world. This paper studies an interesting place finding problem, in which the goal is to correctly identify the interesting places in a city. Important challenges exist in solving this problem: (i) the interestingness of a place is not only related to the number of users who visit it, but also depends upon the travel experience of the visiting users; (ii) the users social connections could directly affect their visiting behavior and the interestingness judgment of a given place. In this paper, we develop a new Social-aware Interesting Place Finding Plus (SIPF+) approach that addresses the above challenges by explicitly incorporating both the users travel experience and social relationship into a rigorous analytical framework. The SIPF+ scheme can find interesting places not typically identified by traditional travel websites (e.g., TripAdvisor, Expedia). We compare our solution with state-of-the-art baselines using two real-world datasets collected from location-based social network services and verified the effectiveness of our approach.

The original Pawlaks rough set approach based on indiscernibility relation (single granularity) has been extended to multigranulation rough set structure in the recent years. Multigranulation rough set approach has become a flouring research direction in rough set theory. This paper considers rough approximation of a fuzzy concept under the framework of multigranulation over two different universes of discourse, i.e., multigranulation fuzzy rough set models over two universes. We present three types of multigranulation fuzzy rough set over two universes by the constructive approach, respectively. Some interesting properties of the proposed models are discussed and also the interrelationships between the proposed models and the existing rough set models are given. We then propose a new approach to a kind of multiple criteria group decision making problem based on multigranulation fuzzy rough set model over two universes. The decision rules and algorithm of the proposed method are given and an example of handling multiple criteria group decision making problem of clothes ranking illustrates this approach. The main contribution of this paper is twofold. One is to establish the multigranulation fuzzy rough set theory over two universes. Another is to try presenting a new approach to multiple criteria group decision making based on multigranulation fuzzy rough set over two universes. The proposed models not only enrich the theory of multigranulation rough set but also make a tentative to provide a new perspective for multiple criteria group decision making with uncertainty.

This paper expands the use of ensemble methods for the prediction of number of faults unlikely the earlier works on ensemble methods that focused on predicting software modules as faulty or non-faulty.This paper investigates the usage of both heterogeneous ensemble methods as well as homogeneous ensemble methods for the prediction of number of faults.We present two linear combination rules and two non-linear combination rules for combining the outputs of the base learners in the ensemble.In addition, we assess the performance of ensemble methods under two different scenarios, intra-release prediction and inter-releases prediction.The experiments are performed over five open-source software systems with their fifteen releases, collected from the PROMISE data repository. Several classification techniques have been investigated and evaluated earlier for the software fault prediction. These techniques have produced different prediction accuracy for the different software systems and none of the technique has always performed consistently better across different domains. On the other hand, software fault prediction using ensemble methods can be very effective, as they take the advantage of each participating technique for the given dataset and try to come up with better prediction results compared to the individual techniques. Many works are available for classifying software modules being faulty or non-faulty using the ensemble methods. These works are only specifying that whether a given software module is faulty or not, but number of faults in that module are not predicted by them. The use of ensemble methods for the prediction of number of faults has not been explored so far. To fulfill this gap, this paper presents ensemble methods for the prediction of number of faults in the given software modules. The experimental study is designed and conducted for five open-source software projects with their fifteen releases, collected from the PROMISE data repository. The results are evaluated under two different scenarios, intra-release prediction and inter-releases prediction. The prediction accuracy of ensemble methods is evaluated using absolute error, relative error, prediction at level l, and measure of completeness performance measures. Results show that the presented ensemble methods yield improved prediction accuracy over the individual fault prediction techniques under consideration. Further, the results are consistent for all the used datasets. The evidences obtained from the prediction at level l and measure of completeness analysis have also confirmed the effectiveness of the proposed ensemble methods for predicting the number of faults.

Social networks with millions of nodes and edges are difficult to visualize and understand. Therefore, approaches to simplify social networks are needed. This paper addresses the problem of pruning social network while not only retaining but also improving its information propagation properties. The paper presents an approach which examines the nodal attribute of a node and develops a criterion to retain a subset of nodes to form a pruned graph of the original social network. To authenticate feasibility of the proposed approach to information propagation process, it is evaluated on small world properties such as average clustering coefficient, diameter, path length, connected components and modularity. The pruned graph, when compared to original social network, shows improvement in small world properties which are essential for information propagation. Results also give a significantly more refined picture of social network, than has been previously highlighted. The efficacy of the pruned graph is demonstrated in the information diffusion process under Independent Cascade (IC) and Linear Threshold (LT) models on various seeding strategies. In all size ranges and across various seeding strategies, the proposed approach performs consistently well in IC model and outperforms other approaches in LT model. Although, the paper discusses the problem with the context of information propagation for viral marketing, the pruned graph generated from the proposed approach is also suitable for any application, where information propagation has to take place reasonably fast and effectively.

The prediction of protein structures directly from amino acid sequences is one of the biggest challenges in computational biology. It can be divided into several independent sub-problems in which protein secondary structure (SS) prediction is fundamental. Many computational methods have been proposed for SS prediction problem. Few of them can model well both the sequence-structure mapping relationship between input protein features and SS, and the interaction relationship among residues which are both important for SS prediction. In this paper, we proposed a deep recurrent encoder–decoder networks called Secondary Structure Recurrent Encoder–Decoder Networks (SSREDNs) to solve this SS prediction problem. Deep architecture and recurrent structures are employed in the SSREDNs to model both the complex nonlinear mapping relationship between input protein features and SS, and the mutual interaction among continuous residues of the protein chain. A series of techniques are also used in this paper to refine the model’s performance. The proposed model is applied to the open dataset CullPDB and CB513. Experimental results demonstrate that our method can improve both Q3 and Q8 accuracy compared with some public available methods. For Q8 prediction problem, it achieves 68.20% and 73.1% accuracy on CB513 and CullPDB dataset in fewer epochs better than the previous state-of-art method.

Differently from Vector-pattern-oriented Classifier Design (VecCD), Matrix-pattern-oriented Classifier Design (MatCD) is expected to manipulate matrix-oriented patterns directly rather than turning them into a vector, and further demonstrated its effectiveness. However, some prior information, such as the local sensitive discriminant information among matrix-oriented patterns, might be neglected by MatCD. To overcome such flaw, a new regularization term named R LSD is adopted into MatCD by taking advantage of Locality Sensitive Discriminant Analysis (LSDA) in this paper. In detail, the objective function of LSDA is modified and transformed into the regularization term R LSD to explore the local sensitive discriminant information among matrix-oriented patterns. In the implementation, R LSD is collaborated with one typical MatCD, whose name is Matrix-pattern-oriented Ho-Kashyap Classifier (MatMHKS), so as to create a new classifier based on local sensitive discriminant information named LSDMatMHKS for short. Finally, comprehensive experiments are designed to validate the effectiveness of LSDMatMHKS. The major contributions of this paper can be concluded as (1) improving the classification performance and the learning ability of MatCD, (2) introducing local sensitive discriminant information into MatCD and extending the application scenario of LSDA, and (3) validating and analyzing the feasibility and effectiveness of R LSD .

Multi-attribute group decision making (MAGDM) has attracted more and more attention in many fields. Correspondingly, a number of usable methods have been proposed for various MAGDM problems, nevertheless, very few research focus on the aggregation techniques of intuitionistic fuzzy information. The aim of this paper is to aggregate decision information into interval-valued intuitionistic fuzzy numbers (IVIFNs) to solve heterogeneous MAGDM problem in which the decision information involves real numbers, interval numbers, triangular fuzzy numbers (TFNs) and trapezoidal fuzzy numbers (TrFNs). There are three issues being addressed in this paper. The first is to propose a new general method to aggregate the attribute value vector into IVIFNs under heterogeneous MAGDM environment utilizing the relative closeness in technique for order preference by similarity to ideal solution (TOPSIS). The second is to construct a multiple objective intuitionistic fuzzy programming model to determine the attribute weights. Borrowing the results of the former two issues, the last is to present a new method to solve heterogeneous MAGDM problem. A comparison analysis with existing method is conducted to demonstrate the advantages of the proposed method. Two examples are provided to verify the practicality and effectiveness of the proposed method.

Time-stamped data are ubiquitous in our daily life, such as twitter data, academic papers and sensor data. Finding clusters and their evolutionary trends in time-stamped data sets are receiving increasing attention from researchers. Most existing methods, however, can only tackle the clustering problem of a data set without time-stamped information which is inherent in almost all the data objects. Actually, not only the performance can be improved by effectively incorporating the time-stamped information in the clustering process on most data sets, but also we can find the evolutionary trends of the clusters with time information. In this paper, we introduce an approach for clustering time-stamped data and discovering the evolutionary trends of the clusters by using Multiple Nonnegative Matrices Factorization (MNMF) with smooth constraint over time. To utilize time-stamped information in the clustering process, an extra object-time matrix is constructed in our proposed method. Then, we jointly factorize multiple feature matrices using smooth constraint to perform the object-time matrix to obtain the clusters and their evolutionary trends. Experimental results on real data sets demonstrate that our proposed approach outperforms the comparative algorithms with respect to Fscore, NMI or Entropy.

There has been a surge in the research of complex network analysis in the recent years. This paper engages with online social network, which is the most popular complex network in the modern world. Network communities help to understand the organization of real world networks. Accordingly, this paper proposes and validates a novel algorithm for overlapping community detection in online social networks. We focus on the stability-plasticity problem in complex networks and attempt to solve it using a Fuzzy Adaptive resonance theory inspired algorithm. The algorithm consists of two stages namely prediction stage and comparison stage. The proposed algorithms make use of network measures such as Edge betweenness, Betweenness centrality, and pair betweenness. The algorithm has been tested and compared with other algorithms using benchmark datasets, artificial datasets and real network datasets. The experimental results obtained were better than other overlapping community detection algorithms. The entropy of the proposed model has been evaluated using Overlapping normalized information, omega index, F-score and the cumulative performance value is 2.42 out of 3, which is better than other community detection algorithm.

Learn a dictionary based on discriminative energy contribution for image classification. A linear classifier is designed in dictionary learning process for efficient classification. The ź2 norm-based dictionary learning benefits for convenient computation. DECDL fulfills the classification task with small number of training samples. Experiments on the face/texture databases verify that DECDL outperforms the state-of-the-art methods. This paper combines the discriminative feature extraction and effective classifier construction into a single framework to learn a structured discriminative dictionary for image classification. Due to the fact that the discriminative signal lie in a low dimensional subspace and can be well represented only via a few atoms of the learned dictionary, this paper addresses the feature extraction via learning a dictionary, whose sub dictionaries preserve correspondence to the class labels, and an optimal linear classifier jointly based on the structure of energy contribution. Based on the discriminative energy contributions, we are searching the discriminative feature for classification rather than reconstructing the data accurately. In addition, with the assumption that the classifier has a specific property which is similar with the dictionary, we learn a classifier to make the dictionary optimal and have a low cost on classifying. Experiment results on the several databases to specific classification tasks are conducted to verify the efficacy of the proposed method compared with the state-of-the-art dictionary learning for classification methods.

In this paper we present a system that automatically builds ordered timelines of events from different written texts in English. The system deals with problems such as automatic event extraction, cross-document temporal relation extraction and cross-document event coreference resolution. Its main characteristic is the application of three different types of knowledge: temporal knowledge, lexical-semantic knowledge and distributional-semantic knowledge, in order to anchor and order the events in the timeline. It has been evaluated within the framework of SemEval 2015. The proposed system improves the current state-of-the-art systems in all measures (up to eight points of F1-score over other systems) and shows a significant advance in the Cross-document event ordering task. This paper has been partially supported by the Spanish government, project TIN2015-65100-R and project TIN2015-65136-C2-2-R.

We extend the community forest model to overlapping community forest model.We propose the definition of overlapping community and disjoint community.We develop a novel overlapping community detection algorithm named CFM.CFM has better performance than MMSB, Louvain method and CPM. Overlapping community detection is the key research work to discover and explore the social networks. A great deal of work has been devoted to detect overlapping communities, but no one can give a clear formula definition of community from the internal structure to the external boundary. More in depth, there are four challenges to existing research works. In this paper, firstly we propose overlapping community forest model and disjoint community forest model based on the community forest model, secondly give a clear formula definition of overlapping community and disjoint community based on the backbone degree and expansion, thirdly propose a novel algorithm to find overlapping communities based on the backbone degree and expansion to resolve the four challenges. This algorithm has better performance than four related algorithms mentioned by this paper in large scale social networks. It works well on American college football, Zachary's Karate Club, Netscience-coauthor, Condensed matter collaborations, LFR etc. data sets.

Decision-making is the most important business activity and becomes more complex in the current big-data situation. Most organizational decision-making is made in a group and has the data analytics function to seek specific answers for specific purposes. Multi-attribute group decision-making (MAGDM) methods provide effective support to decision groups by evaluating and integrating individual group members opining. However, current MAGDM methods often suffer from the problem of opining data and decision environment uncertainty, which is particularly severe in a large decision group or a newly decision problem. As a solution, rough sets and fuzzy sets have been applied in MAGDM to deal with data and decision process uncertainties. Although a lot of efforts have been made in applying rough sets and fuzzy sets to deal with data and decision process uncertainties the area, the disadvantage of rough set models in classification accuracy when similar class as basic knowledge granularity has not been well solved yet. This paper aims to solve this problem by introducing a new concept-maximal consistent block (MCB) and multi-granulation decision-theoretic rough set (MCB-MDTRS) models. Firstly, it establishes a binary tolerance relation on the universe of discourse, defines generalized MCBs, and introduces pessimistic DTRS based on MCBs. Secondly, it extends objective set to a fuzzy environment and proposes four MCB-MDTRFS models with consideration of the weight of each attribute. The steps of the proposed fuzzy MAGDM methods are carefully described in detail. Different from existing fuzzy MAGDM methods, the weight of each attribute is considered in the determination of the positive ideal decision objective and negative ideal decision objective in this paper, named by qualified and unqualified (fuzzy) sets. We solve the vague problem of individual preference evaluation and the uncertainty of setting ideal decision objectives by using the advantages of fuzzy set and rough set theory. Finally, it takes emergency plan selection as a case study to analyze the effectiveness of our methods and compare with other fuzzy MAGDM methods. Our methods outperform the selection of basic knowledge granularity, the determination of ideal decision objective sets and the ranking method, so as to increase the reliability and accuracy of ranking evaluation index.

Class imbalance is an active research area in the machine learning community. However, existing and recent literature showed that class overlap had a higher negative impact on the performance of learning algorithms. This paper provides detailed critical discussion and objective evaluation of class overlap in the context of imbalanced data and its impact on classification accuracy. First, we present a thorough experimental comparison of class overlap and class imbalance. Unlike previous work, our experiment was carried out on the full scale of class overlap and an extreme range of class imbalance degrees. Second, we provide an in-depth critical technical review of existing approaches to handle imbalanced datasets. Existing solutions from selective literature are critically reviewed and categorised as class distribution-based and class overlap-based methods. Emerging techniques and the latest development in this area are also discussed in detail. Experimental results in this paper are consistent with existing literature and show clearly that the performance of the learning algorithm deteriorates across varying degrees of class overlap whereas class imbalance does not always have an effect. The review emphasises the need for further research towards handling class overlap in imbalanced datasets to effectively improve learning algorithms’ performance.

Microblogging is a popular online social network (OSN), which facilitates users to obtain and share news and information. Nevertheless, it is filled with a huge number of social bots that significantly disrupt the normal order of OSNs. Sina Weibo, one of the most popular Chinese OSNs in the world, is also seriously affected by social bots. With the growing development of social bots in Sina Weibo, they are increasingly indistinguishable from normal users, which presents more huge challenges in detecting social bots. Firstly, it is difficult to extract the features of social bots completely. Secondly, large-scale data collection and labeling of user data are extremely hard. Thirdly, the performance of classical classification approaches applied to social bot detection is not good enough. Therefore, this paper proposes a novel framework for detecting social bots in Sina Weibo based on deep neural networks and active learning (DABot). Specifically, 30 features from four categories, namely metadata-based, interaction-based, content-based, and timing-based are extracted to distinguish between social bots and normal users. Nine of these features are completely new features proposed in this paper. Moreover, active learning is employed to efficiently expand the labeled data. Then, a new deep neural network model called RGA is built to implement the detection of social bots, which makes use of a residual network (ResNet), a bidirectional gated recurrent unit (BiGRU), and an attention mechanism. After performance evaluation, the results show that DABot is more effective than the state-of-the-art baselines with the accuracy of 0.9887.

In recent years, context information has been widely used in recommender systems. Tensor factorization is an effective method to process high-dimensional information. However, data sparsity is more serious in tensor factorization, and it is difficult to build a more accurate recommender system only based on user–item–context interaction information. Making full use of user’s social information and implicit feedback can alleviate this problem. In this paper, we propose a new tensor factorization model named TrustTF, which mainly works as follows: (1) Using user’s social trust information and implicit feedback to extend the bias tensor factorization (BiasTF), effectively alleviate data sparsity problem and improve the recommendation accuracy; (2) Dividing user’s trust relationship into unilateral trust and mutual trust, which makes better use of user’s social information. To our knowledge, this is the first work to consider the effects of both user trust and implicit feedback on the basis of the BiasTF model. The experimental results in two real-world data sets demonstrate that the TrustTF proposed in this paper can achieve higher accuracy than BiasTF and other social recommendation methods.

This paper focuses on the problem of student knowledge diagnosis that is a basic task of realizing personalized education. Most traditional methods rely on the question-concept matrix empirically designed by experts. However, the expert concepts are expensive and inter-overlapping in their constructions, leading to ambiguous explanations. With the intuition that each student can master a part of the knowledge involved in all questions, in this paper, we propose a novel learning-based model for student knowledge diagnosis, dubbed Meta-knowledge Dictionary Learning (metaDL). MetaDL aims to learn a meta-knowledge dictionary from student responses, where any knowledge entity (e.g., student, question or expert concept) is a linear combination of a few atoms in the meta-knowledge dictionary. The resultant problem could be effectively solved by developing the alternating direction method of multipliers. This study has three innovations: learning independent meta-knowledges instead of traditional complex concepts, sparely representing knowledge entity instead of densely weighted representation, and interpreting expert concepts with the resulting meta-knowledges. For evaluation, the diagnosis results from metaDL are used to group students and predict responses on two public datasets and a private dataset from our institution. The experiment results show that metaDL delivers an effective student knowledge diagnosis and then results in good performances on the two applications in comparison with other methods. This technique could provide significant insights into student’s knowledge state and facilitate the progress on personalized education.

Crowdsourcing services provide an easy means of acquiring labeled training data for supervised learning. However, the labels provided by a single crowd worker are often unreliable. Repeated labeling can be used to solve this problem. After multiple labels have been acquired by repeated labeling for each instance, in general consensus methods are used to obtain the integrated labels of instances. Although consensus methods are effective in practice, it cannot be denied that a level of noise still exists in the set of integrated labels. In this study, an attempt was made to employ noise filters to delete the noise in integrated labels, and consequently, enhance the training data and model quality. In fact, noise handling is a relatively mature field in the machine learning community, and many noise filters for deleting label noise have been presented in the past. However, to the best of our knowledge, in very few studies was noise filtering used to improve crowdsourcing learning. Therefore, in this study we empirically investigated the performance of noise filters in terms of improving crowdsourcing learning. Thus, in this paper some existing noise filters presented in previous papers are reviewed and their experimental application to crowdsourcing learning tasks is described. Experimental results based on 14 benchmark UCI data sets and three real-world data sets show that these noise filters can significantly reduce the noise level in integrated labels and thereby considerably enhance the performance of target classifiers.

This paper deals with label compression of multi-label classification.It is the first paper considering outliers in label compression.Outliers in the feature space are taken into account.Irregular label correlations can also be thought as outliers.This paper tackles this problem by using l2,1-norm. Label compression (LC) is an effective strategy to reduce time cost and improve classification performance simultaneously for multi-label classification. One main limitation of existing LC methods is that they are prone to outliers. Here outliers include outliers in the feature space and outliers in the label space. Outliers in the feature space are obtained due to data acquisition devices. Outliers in the label space refer to label vectors that are inconsistent with the regular label correlations. In this paper, we propose a new LC method, termed robust label compression (RLC), based on l2,1-norm to deal with outliers in the feature space and label space. The objective function of RLC consists of two losses: the encoding loss to measure the compression error and the dependence loss to measure the relevance between the instances and the obtained code vectors after compressing the label vectors. To achieve robustness to outliers, we utilize the l2,1-norm on both losses. We propose an efficient optimization algorithm for it and present theoretical analysis. Experiments across six data sets validate the superiority of our proposed method to state-of-art LC methods for multi-label classification.

In this paper we propose a new approach to learning fuzzy cognitive maps (FCMs) as a predictive model for time series forecasting. The first contribution of this paper is the dynamic optimization of the FCM structure, i.e., we propose to select concepts involved in the FCM model before every prediction is made. In addition, the FCM transformation function together with the corresponding parameters are proposed to be optimized dynamically. Finally, the FCM weights are learned. In this way, the entire FCM model is learned in a completely new manner, i.e., it is continuously adapted to the current local characteristics of the forecasted time series. To optimize all of the aforementioned elements, we apply and compare 5 different population-based algorithms: genetic, particle swarm optimization, simulated annealing, artificial bee colony and differential evolution. For the evaluation of the proposed approach we use 11 publicly available data sets. The results of comparative experiments provide evidence that our approach offers a competitive forecasting method that outperforms many state-of-the-art forecasting models. We recommend to use our FCM-based approach for the forecasting of time series that are linear and tend to be trend stationary.

This paper focuses on an anomaly detection method that uses a baseline model describing the normal behavior of network traffic as the basis for comparison with the audit network traffic. In the anomaly detection method, an alarm is raised if a pattern in the current network traffic deviates from the baseline model. The baseline model is often trained using normal traffic data extracted from traffic data for which all instances (i.e., packets) are manually labeled by human experts in advance as either normal or anomalous. However, since humans are fallible, some errors are inevitable in labeling traffic data. Therefore, in this paper, we propose an anomaly detection method that is tolerant to human errors in labeling traffic data. The fundamental idea behind the proposed method is to take advantage of the lossy nature of packet sampling for the purpose of correcting/preventing human errors in labeling traffic data. By using real traffic traces, we show that the proposed method can better detect anomalies regarding TCP SYN packets than the method that relies only on human labeling.

International audience; This paper introduces a knowledge-based system which grounds the detection of the user's likes and dislikes on the topic structure of the conversation. The targeted study is set in a human-agent interaction with the aim to help the creation of dialogue strategies of an agent based on the user's interests. In this paper, we first describe the system based on linguistic resources such as lexicons, dependency grammars and dialogue information provided by the dialogue system. Second, we explain how the system merges its outputs at the end of each topic sequence. Finally, we present an evaluation of both the linguistic rules and the merging process. The system enables a better identification of the target of the user's likes and dislikes and provides a synthetic representation of the user's interests.

Top-k high utility itemset mining is the process of discovering the k itemsets having the highest utilities in a transactional database. In recent years, several algorithms have been proposed for this task. However, it remains very expensive both in terms of runtime and memory consumption. The reason is that current algorithms often generate a huge amount of candidate itemsets and are unable to prune the search space effectively. In this paper, we address this issue by proposing a novel algorithm named kHMC to discover the top-k high utility itemsets more efficiently. Unlike several algorithms for top-k high utility itemset mining, kHMC discovers high utility itemsets using a single phase. Furthermore, it employs three strategies named RIU, CUD, and COV to raise its internal minimum utility threshold effectively, and thus reduce the search space. The COV strategy introduces a novel concept of coverage. The concept of coverage can be employed to prune the search space in high utility itemset mining, or to raise the threshold in top-k high utility itemset mining, as proposed in this paper. Furthermore, kHMC relies on a novel co-occurrence pruning technique named EUCPT to avoid performing costly join operations for calculating the utilities of itemsets. Moreover, a novel pruning strategy named TEP is proposed for reducing the search space. To evaluate the performance of the proposed algorithm, extensive experiments have been conducted on six datasets having various characteristics. Results show that the proposed algorithm outperforms the state-of-the-art TKO and REPT algorithms for top-k high utility itemset mining both in terms of memory consumption and runtime.