Search

In this paper, impulse noise removal problem is formulated as an unconstrained optimization problem with smooth objective function. It can be solved by conjugate gradient methods with desired properties (low memory and strong global convergence) in high dimensions. Accordingly, a family of the Polak-Ribi`ere-Polyak (PRP) conjugate gradient directions is constructed for which the descent condition holds. In other words, we introduce four improved versions of PRP method three of which are based on a regularization and one of which is the combination of Fletcher-Reeves and PRP conjugate gradient parameters. Using several images, it is shown that the new methods are very robust and efficient in comparison with other competitive methods for impulse noise removal, especially in terms of the peak signal to noise ratio (PSNR). Mathematics Subject Classification (2000) 90C30a . 90C25 . 90C90 . 68U10 . 03D15

In this paper, impulse noise removal problem is formulated as an unconstrained optimization problem with smooth objective function. It can be solved by conjugate gradient methods with desired properties (low memory and strong global convergence) in high dimensions. Accordingly, a family of the Polyak-Ribi`ere conjugate gradient directions is constructed for which the descent condition holds. Using several images, it is shown that the new methods are very robust and efficient in comparison with other competitive methods for impulse noise removal, especially in terms of the peak signal to noise ratio (PSNR). Mathematics Subject Classification (2000) 90C30a · 90C25 · 90C90 · 68U10 · 03D15

One of the main steps in the data lifecycle is to publish it for data analysts to discover hidden patterns. But, data publishing may lead to unwanted disclosure of personal information and cause privacy problems. Data anonymization techniques preserve privacy models to prevent the disclosure of individuals’ private information in published data. In this paper, a distributed in-memory method is proposed on the Apache Spark framework to preserve the l-diversity privacy model. This method anonymizes large-scale data in a three-phase process, which includes, seed selection, data clustering for $$\ell$$ -diversity, and finalizing phase. In this method, a hierarchical kmeans-based data clustering algorithm has been designed for data anonymization. One of the major challenges of anonymization methods is to establish a better trade-off between data utility and privacy. Therefore, for calculating the distance between records and forming more cohesive ldiverse-clusters, the authors have designed two Manhattan-based and Euclidean-based distance functions to satisfy the requirements of the l-diversity model. Given the 100-fold speed of the Spark compared to MapReduce, the proposed method is presented using in-memory RDD programming in Apache Spark, to address the runtime, scalability, and performance in large-scale data anonymization as it exists in the previous MapReduce-based algorithms. Our method provides general knowledge to use parallel in-memory computation of Spark in big data anonymization. In experiments, this method has obtained lower information loss and loses about 1% to 2% accuracy and FMeasure criteria; therefore, it establishes a better trade-off than the state-of-the-art MapReduce-based Mondrian methods

To offer an appropriate recommendation to customers in recommender systems, the issue of clustering and separating users with different tastes from the rest of people is of significant importance. The MkMeans + + algorithm is a technique for clustering and separating users in collaborative filtering systems. This algorithm utilizes a specific procedure for selecting the initial centroids of the clusters and has a better function compared with its similar algorithms such as kMeans + + . In this paper, MkMeans + + algorithm is combined with Firefly, Cuckoo, and Krill algorithms and new algorithms called FireflyMkMeans + + , CuckooMkMeans + + , and KrillMkMeans + + are introduced in order to specify the optimal centroid of the cluster, better separate users, and avoid local optimals. In the proposed hybrid clustering approach, the initial population of firefly, cuckoo, and krill algorithms is initialized through the solutions generated by MkMeans + + algorithm, and it makes use of the benefits of MkMeans + + as well as firefly, cuckoo, and krill algorithms. Results and implementations on both MovieLens and FilmTrust datasets indicate that the proposed algorithms can perform better than their similar algorithms in clustering and separating users with different tastes (graysheep users), and enhance the quality of clusters and the accuracy of recommendations for users with similar tastes (white users).

For the extraction of useful patterns, the collected data should be distributed to and shared with analyzers. This, however, creates problems and challenges for the individual with respect to their privacy and identity. In this paper, the Mondrian multidimensional anonymization method was developed and improved for satisfaction of the l-diversity privacy model, and it has been presented in a distributed fashion within the Apache Spark framework. Since one of the major challenges in data privacy is the tradeoff between privacy and data utility, the presented method focuses on information loss and classifier evaluation criteria. Therefore, the cut dimension was selected using the coefficient of variation and information gain criteria, and the cut points were chosen dynamically, which led to a decrease in the information loss parameter and an improvement in the classifier performance evaluation criteria such as accuracy and FMeasure compared to the previous algorithms in the literature. The processing speed is 100 times higher in Spark than in the Hadoop framework. Consequently, the proposed method was presented in a distributed fashion based on RDDs programming within Apache Spark framework. This will resolve the problem of speed in large-scale data anonymization as it exists in the previous Hadoop-based algorithms. The results of the experiments performed on the numerical datasets demonstrate the improvements made by the proposed method.

One of the most important resources used in today's world is image. Medical images can play an essential role in helping diagnose diseases. Doctors and specialists use medical images to diagnose brain diseases. Convolution neural networks are among the most important deep learning methods in which several layers are trained powerfully. The proposed method is a deep parallel convolution neural network model consisting of AlexNet and VGGNet networks. The structure of the layers is different for the two types of networks. The features are combined at one point and then categorized using the softmax function. To test the network, we used different criteria with images in the database. In clinical diagnosis, generalizability means prediction for people from whom we have no observation. On this account, we did not include individuals' observations in training set in the test set. Compared to the existing models, the proposed model results show that our network has achieved better results. The best result for the database was FIGSHARE, which achieved 99.14% accuracy on binary class and 98.78% on multi-class, which was much better than other SVM models. We tested the best available result with the existing database and compared the results with the proposed model, which was the best model answer for all evaluation criteria. These results suggest that the proposed model can be used as an effective decision support tool for radiologists in medical diagnosis.

Ever increasing popularity social networks has attracted many companies to use them for viral marketing. Identification of influential users for spreading news and marketing is a major challenge in viral marketing. In most of the existing influence maximization approaches, the topological locations of nodes on the network has been considered as a criterion to determine their influentiality. They have mainly neglected users’ interest in the marketing message. Although a number of existing works have considered interests of users, they have not used any criterion to specify the interest. This manuscript proposes a novel criterion to measure the interest of users in the marketing messages. We then propose a novel algorithm to obtain the set of the most influential users. Experimental results on real-world and synthetic networks reveal effectiveness of the proposed method as compared to the existing state-of-the-art influence maximization algorithms.

Social networks have significant role in distribution of ideas and advertisement. Discovering the most influential nodes has been a hot topic in the field of social networks analysis and mining. This manuscript proposes novel algorithms for this purpose based on neighborhood diversity. We introduce two new influential node ranking algorithms that use diversity of the neighbors of each node in order to obtain its ranking value. They are applied on a number of real-world networks and compared with state-of-the-art algorithms. Our experimental results reveal effectiveness of the proposed algorithms.

In recent research, the importance of determining social network users’ spreading influence and ranking them has attracted plenty of attention. H-index is one of the methods that have been presented for this purpose, and determines the spreading capability of a node based on the degrees of its neighbors. In this method, part of the information on the neighbors is disregarded, which reduces ranking accuracy. In this paper, a measure is presented for specification of the centrality of nodes through extension of the H-index notion. The results of experimentation over real-world and artificial datasets demonstrate that the proposed measure exhibits higher accuracy and efficiency than in the other compared methods.

Identification of influential users for news and messages propagation constitutes one of the most important topics in analysis of social networks. The success of the spreading process in these networks depends on the mechanism for selection and specification of the influential users. Beside selection of influential users, the distances between selected users should be considered in this mechanism to ensure minimum overlap and maximum coverage of a wider area of the network. Simultaneous meeting of the two objectives may be contradictory. In this paper, we propose a new approach for selection of the set of influential users using the Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS) method that seeks to present a solution by considering the above two objectives. The simulation results over real-world and artificial networks demonstrate that the set selected by the proposed approach exhibits greater spread of influence than those selected by the traditional approaches.

Purpose This paper aims to analyze the role of influential nodes on other users on Facebook social media sites by social and behavioral characteristics of users. Hence, a new centrality for user is defined, applying susceptible-infected recovered (SIR) model to identify influence of users. Results show that the combination of behavioral and social characteristics would be determined the most influential users that influence majority of nodes on social networks. Design/methodology/approach In this paper, the authors define a new centrality for users, considering node status and behaviors. Thus, this node has a high level of influence. Node social status includes node degree, clustering coefficient and average neighbors’ node, and social status of node refers to user activities on Facebook social media website such as sending posts and receiving likes from other users. According to social status and user activity, the new centrality is defined. Finally, through the SIR model, the authors explore infection power of nodes and their influences of other node in the network. Findings Results show that the proposed centrality is more effective than other centrality approaches, infecting more nodes in social network. Another significant point in this research is that users who have high social status and activities on Facebook are more influential than users who have only high social status on the Facebook social media. Originality/value The influence of user on others in social media includes two key factors. The first factor is user social status such as node degree and clustering coefficient in social media graph and the second factor is related to user social activities in social media sites. Most centralities focused on node social status without considering node behavior. This paper analyzes the role of influential nodes on other users on Facebook social media site by social and behavioral characteristics of users.

Identification of influential nodes in complex networks is a significant problem.We propose a centrally measure to identify nodes location in complex networks.Proposed measure is applied to rank the influential nodes.Experiments show better results as compared to the state-of-the-art methods. Due to the rapid extension of social networks in recent years, a new potential has emerged for global spreading of messages and effective broadcasting of news. Identification of influential nodes within a network is now seen as a key factor for bringing this potential into action. k-shell is a measure for detection of node influence, and has already been used in some successful algorithms in this field. However, k-shell does not provide enough information about the topological positions of the nodes, and the present paper seeks to present a special hierarchical measure for dealing with this issue. Along with introduction of the above measure, it is shown that it can be used for detecting and ranking node influence. The experiments done on real-world and artificial networks demonstrate that the proposed approach can rank the influence of nodes more accurately than other approaches.

Identification and ranking of influential users in social networks for the sake of news spreading and advertising has recently become an attractive field of research. Given the large number of users in social networks and also the various relations that exist among them, providing an effective method to identify influential users has been gradually considered as an essential factor. In most of the already-provided methods, those users who are located in an appropriate structural position of the network are regarded as influential users. These methods do not usually pay attention to the interactions among users, and also consider those relations as being binary in nature. This paper, therefore, proposes a new method to identify influential users in a social network by considering those interactions that exist among the users. Since users tend to act within the frame of communities, the network is initially divided into different communities. Then the amount of interaction among users is used as a parameter to set the weight of relations existing within the network. Afterward, by determining the neighbors’ role for each user, a two-level method is proposed for both detecting users’ influence and also ranking them. Simulation and experimental results on twitter data shows that those users who are selected by the proposed method, comparing to other existing ones, are distributed in a more appropriate distance. Moreover, the proposed method outperforms the other ones in terms of both the influential speed and capacity of the users it selects.

Measurement of the spreading capability of nodes has been one of the most attractive challenges in the field of social networks. Because of the huge number of nodes in a network, it has appealed to many researchers to find an accurate measure which can potentially detect the spreading capability and rankings of nodes. Most of the available methods determine the spreading capability of nodes based on their topological locations. In this paper, however, we have proposed a new measure based on the basic notions in information theory to detect the spreading capability of nodes in networks on the basis of their topological information. The simulation and experimental results of investigating real-world and artificial networks show that the proposed measure is more accurate and efficient than the similar ones.

Link prediction is an important issue in complex network analysis and mining. Given the structure of a network, a link prediction algorithm obtains the probability that a link is established between two non-adjacent nodes in the future snapshots of the network. Many of the available link prediction methods are based on common neighborhood. A problem with these methods is that if two nodes do not have any common neighbors, they always predict a chance of zero for establishment of a link between them; however, such nodes have been shown to establish links in some real systems. Another issue with these measures is that they often disregard the connection direction. Here, we propose a novel measure based on common neighborhood that resolves the above issues. The proposed measures are applied on three benchmark networks in both unsupervised and supervised learning modes. Our experiments show the superior performance of the proposed measures over that of the state-of-the-art link prediction methods.

Identifying high spreading power nodes is an interesting problem in social networks. Finding super spreader nodes becomes an arduous task when the nodes appear in large numbers, and the number of existing links becomes enormous among them. One of the methods that is used for identifying the nodes is to rank them based on k-shell decomposition. Nevertheless, one of the disadvantages of this method is that it assigns the same rank to the nodes of a shell. Another disadvantage of this method is that only one indicator is fairly used to rank the nodes. k-Shell is an approach that is used for ranking separate spreaders, yet it does not have enough efficiency when a group of nodes with maximum spreading needs to be selected; therefore, this method, alone, does not have enough efficiency. Accordingly, in this study a hybrid method is presented to identify the super spreaders based on k-shell measure. Afterwards, a suitable method is presented to select a group of superior nodes in order to maximize the spread of influence. Experimental results on seven complex networks show that our proposed methods outperforms other well-known measures and represents comparatively more accurate performance in identifying the super spreader nodes.

The aim of this research is to predict fault in the process of producing important Android applications using data mining techniques. Predictable models must not only be correct in predicting fault, but also must be understandable, which needs the user to understand the motivation of the model prediction. Unfortunately, understandability of the fault types is ignored in order to achieve the predictable efficiency of the fault prediction models. In order to solve this problem, some trees are extracted from the random forests and support vector machines for the logistic regression and the rule extraction algorithm are used; also, NASA MDP data are used for extracting the model. The method of evaluating the prediction of software fault is the use of ALPA algorithm and extraction of random trees (RF) and (Logistic) regression for weka software. In the method of creating trees, (REPTree and C4.5) and black box model (Logistic, RF) are used. The results show that the trees extracted from the black box models discuss the prediction way of black box which is more understandable as well as more correct than the trees from direct work on the data.

Advertisement over social networks is critical for many businesses, and selecting the initial set of influential nodes for which the advertising message is passed, is regarded as an important issue in this regard. Although various measures have been proposed for specifying the influentiality of a set, it is affected by different factors. In this paper, a multi-objective function is first defined as an influentiality measure, and finding such an initial is framed as an optimization problem. Then, using artificial bee colony optimization two approaches are proposed to solve the problem. In the first approach, influentiality of nodes is only taken into account, while in the second method a budget constraint is also considered. Different experiments on real networks are conducted to evaluate the proposed methods, where the obtained results show their outperformance over state-of-the-art influence maximization methods.

In the recommender systems, the users evaluate and rate data items and assign quantitative indices to them; consequently, a comprehensive database called rating matrix is created. In the rating matrix, the separation of the gray sheep users is of crucial importance since these users increase the error rate of the recommendations to the white users. The proposed approach involved three main steps. In the first step, we introduced a new algorithm called MKMeans++ by extracting and combining some features of the rating matrix as well as applying these combinations to the initial centers of the KMeans clustering, in which Power Item and Power Weight features were added to the new combinations. These features indicated the superiority of the proposed algorithm over the KMeans++ algorithm. In the second step, we probed into the relationship between features of the rating matrix and the gray sheep users, the results of which revealed that the gray sheep users can be identifiable only through the Distance feature. In the third step, we started separating the gray sheep users by using MKMeans++ algorithm and applying the Distance feature. Experimental results on the MovieLens and FilmTrust datasets show that the proposed algorithm outperforms existing methods in terms of the MAE accuracy and Converge. In the proposed algorithm, on average, the amount of the MAE improvement for MovieLens and FilmTrust datasets were 0.15 and 0.24, respectively, while also increasing the quality of clusters and the precision of the recommender system.

Finding the most influential nodes in social networks has significant applications. A number of methods have been recently proposed to estimate influentiality of nodes based on their structural location in the network. It has been shown that the number of neighbors shared by a node and its neighbors accounts for determining its influence. In this paper, an improved cluster rank approach is presented that takes into account common hierarchy of nodes and their neighborhood set. A number of experiments are conducted on synthetic and real networks to reveal effectiveness of the proposed ranking approach. We also consider ground-truth influence ranking based on Susceptible–Infected–Recovered model, on which performance of the proposed ranking algorithm is verified. The experiments show that the proposed method outperforms state-of-the-art algorithms.

A challenging issue in viral marketing is to effectively identify a set of influential users. By sending the advertising messages to this set, one can reach out the largest area of the network. In this paper, we formulate the influence maximization problem as an optimization problem with cost functions as the influentiality of the nodes and the distance between them. Maximizing the distance between the seed nodes guarantees reaching to different parts of the network. We use gray wolf optimization algorithm to solve the problem. Our experimental results on three real-world networks show that proposed method outperforms state-of-the-art influence maximization algorithms. Furthermore, it has lower computational time than other meta-heuristic methods.

The prevalence of social networks amongst people has become an inevitable issue. At the same time, social networks have widely been used for commercial purposes. As a result, in order to sell the products, social networks have been equipped with various recommender systems that provide the users with commercial offers that are appropriate for their behavior. The accuracy of the recommender systems in providing offers to the users and the number of offers accepted by the users are crucial issues. In the present study, a recommender system was designed that operates based on the users' behavior on Facebook and in two phases offers the users to buy their favorite products. In the first phase, the users' behavior is investigated and based on their interest they are offered to buy some products. In the second phase, the recommender system uses data mining techniques and provides the users with offers that are relevant to their previous purchase. The data of the study are factual and the results are valid. Moreover, the results indicate that the designed recommender system is highly accurate in providing offers to the users.

Social networks have turned into a popular medium for diffusion of information, providing connections between people through access to several networks and shared personal opinions, thoughts, information, and experiences. A web-based social network consists of users who are increasingly concerned with protection of their privacy, which is considered as an important concern about user privacy due to the nature of social networks and privacy protection. Therefore, every network user prefers to identify people trusted by him and to communicate with them in order not to be abused by untrusted people. In this paper, we seek to predict trust links by utilizing the most important features of each user on a social network, so that the users can pursue their everyday purposes without much concern. In order to provide trust in social networks, we made use of several features of the users at the same time, the most important of which include the extent to which they trust each other, the amount of similarity between their trust, and each one’s reputation. Using these measures, a trust path was created on the network for each connected user, compared with four common methods for verification of its truth. The results demonstrate that the proposed method has been capable of outperforming the Katz, h Trust, TP, and RS methods in terms of effectiveness, efficiency, and strength, and can more accurately present a more reliable path within an acceptable runtime.

Recently an increasing amount of research is devoted to the question of how the most influential nodes (seeds) can be found effectively in a complex network. There are a number of measures proposed for this purpose, for instance, high-degree centrality measure reflects the importance of the network topology and has a reasonable runtime performance to find a set of nodes with highest degree, but they do not have a satisfactory dissemination potentiality in the network due to having many common neighbors ($\mbox{CN}^{(1)}$) and common neighbors of neighbors ($\mbox{CN}^{(2)}$). This flaw holds in other measures as well. In this paper, we compare high-degree centrality measure with other well-known measures using ten datasets in order to find a proportion for the common seeds in the seed sets obtained by them. We, thereof, propose an improved high-degree centrality measure (named DegreeDistance) and improve it to enhance accuracy in two phases, FIDD and SIDD, by putting a threshold on the number of common neighbors of already-selected seed nodes and a non-seed node which is under investigation to be selected as a seed as well as considering the influence score of seed nodes directly or through their common neighbors over the non-seed node. To evaluate the accuracy and runtime performance of DegreeDistance, FIDD, and SIDD, they are applied to eight large-scale networks and it finally turns out that SIDD dramatically outperforms other well-known measures and evinces comparatively more accurate performance in identifying the most influential nodes., Comment: 17 pages, 8 figures, 5 tables . . . accepted for publication in Physica A: Statistical Mechanics and its Applications

As the Internet has found its place within different sectors of society, the use of various social networks is increasing and that is why they are utilized for business purposes. Most commercial companies need to do marketing in social networks in order to introduce their goods and/or services. In marketing, appropriate and accurate advertisement of goods based on needs of users of these social networks is required to sell the goods of different companies and manufacturing units in effective ways. In most cases, however, advertising and marketing methods in social networks are not correct and the resulted advertisements are boring and tiring to users and they might ignore them without noticing. Such advertisements will be considered as spam and will be annoying to network users. This is taken into consideration in the present study where users' interests, attitudes, and behavior on Facebook are specified through data mining techniques, based on which content marketing is conducted. The present study is conducted on the social network of Facebook, where content marketing, a new form of marketing, is utilized and instead of introducing the goods, the contents of different goods are presented. The data utilized in the study are actual and related to Facebook users, and therefore, the results can be generalized to other networks.