کنفرانس بین المللی وب پژوهی
سال:1398 | دوره:6 |تعداد مقالات:84

Product reviews are one of the most important types of user-generated contents that are becoming more and more available. These reviews are valuable sources of knowledge for users who want to make purchasing decisions and for producers who want to improve their products and services. However, not all product reviews are equally helpful and this makes the process of finding helpful reviews among the massive number of similar reviews very challenging. To address this problem, automatic review helpfulness prediction systems are designed to classify reviews according to their content. In this study, a deep model is proposed to utilize content-based, semantic, sentiment, and metadata features of reviews for predicting review helpfulness. In the proposed method, convolution layer is used for learning feature maps and gated recurrent units are employed for exploiting sequential context. The results of comparing the proposed method with five traditional learning methods and two deep models trained on the same types of features shows that the proposed method outperforms other methods by 4% and 2% in terms of F1-measure and accuracy. Moreover, results reveal that both textual and metadata features are important in detecting helpful reviews. The findings of this study may help online retailers to efficiently rank the product reviews.

The data being generated today is massive in terms of volume, velocity, and variety. It is a great challenge to derive knowledge from data in this condition. Researchers, therefore have proposed ways to deal with this challenge. frequent itemset mining is one of the proposed ways to distinguish itemsets inside the vast amount of data to aid the operation of a variety of pursuits and businesses, a process termed ‘ Association Rule Mining’ . However, there are a variety of works done in this area. The introduction of different algorithms, frameworks, and applications throughout the recent decade has produced many interesting approaches. One of the algorithms in this area is the Apriori algorithm. It is a simple yet powerful algorithm. However, the original Apriori is not suitable for big data and due to this reason, researchers have attempted to introduce ways and schemes to adapt it to this new age of data. Because of the number of efforts in this area, having a bird’ s eye view of the past works is of value. This review aims to present an insight into the works done in the intersection of two matters: big data and the Apriori algorithm. It is concerned with Aprioribased algorithms presented in the recent decade with a focus on the three popular big data platforms: Apache Hadoop, Spark, and Flink. Also, major points of each approach and solution is presented. A conclusion in the end summarizes the points discussed in this paper.

Web data such as documents, images, and videos are examples of large matrices. To deal with such matrices, one may use matrix decomposition techniques. As such, CUR matrix decomposition is an important approximation technique for high-dimensional data. It approximates a data matrix by selecting a few of its rows and columns. However, a problem faced by most CUR decomposition matrix methods is that they ignore the correlation among columns (rows), which gives them lesser chance to be selected; even though, they might be appropriate candidates for basis vectors. In this paper, a novel CUR matrix decomposition method is proposed, in which calculation of the correlation, boosts the chance of selecting such columns (rows). Experimental results indicate that in comparison with other methods, this one has had higher accuracy in matrix approximation.

Service-oriented architecture (SOA) is a successful approach to conquer some of the complexities of distributed systems. There are many isolated and composite services being used based-on this approach. The composition of different service types creates a complex network used in various service compositions. However, the analysis of these networks, such as robustness analysis, have not received appropriate attention by the research community yet. In this paper, the web service networks are analyzed considering the robustness of these networks against random and targeted attacks. The results indicate that the robustness of these networks is low against both random and targeted attacks.

Due to the increasing use of users' opinions in various domains on social networks and the importance of these opinions, their accuracy is very important, but unknown persons may use fake comments to promote or discredit products, services, Organizations or peoples. Since it is difficult and even impossible to identify only on through read, find data will be difficult to design and evaluate the algorithms for the identification of opinion spam too. Due to the challenge explained, the present paper, by innovating in the combination of opinion content, metadata and entity information, generates a set of data features and for the first time at the document and sentence level, recognizes opinion spam in Persian. In the following, the identification of opinion spam as a classification problem is introduced with two fake and non-fake categories and is modeled with six supervised learning methods. To evaluate the results, the Confusion matrix of each method is constructed and after calculating the precision, recall and accuracy and comparing the values, the best and most accurate classification will be introduced in identifying opinion spam.

Text analysis is a method for extracting knowledge from text. Memory and time limitations in processing big data is crucial due to data sources distributed in web, search engines and socials network sites. In addition, due to automatizing search process, summarizing and finding the interests of users, immediate classification of various texts in a streaming manner has gained attention in industrial and scientific fields. Hierarchical classification of text is among common issues which is simply possible in traditional methods using bag of words; however, while talking about big data and when there are a lot of labels of classes, employing traditional methods will not meet the needs of societies. With the improvement of data in internet and social networks, more powerful methods are needed which can classify the data closely and immediately. Through abstraction in textual data, deep learning can deal with these challenges. In this paper a deep learning method will be introduced which is based on hierarchical classification (HAN) named HAN-MODI and which can classify texts from social networks and web sites with an accuracy of 98. 81% at the real time bilingually in English and Farsi. This paper also shows that this complex network with three modules word, sentence and document can work better at word level and there is no need to know syntactic or semantics structure of language. The novelty of the proposed method is adding a third level to the hierarchical structure for general detection and for more exact detection of the class. In addition, classification using this method will be multi-level classification and finally with a change in HAN, this method can be used with Farsi texts. Model improvement is done by adding a new layer above the architecture HAN. We called it as segmentation of sentences into expressions Bag of Sentences and added a dynamicity window in any stage that applied attention mechanism simultaneously.

We are living in the data age. Communications over scientific networks creates new opportunities for researchers who aim to discover the hidden pattern in these huge repositories. This study utilizes network science to create collaboration network of Iranian Scientific Institutions. A modularity-based approach applied to find network communities. To reach a big picture of science production flow, analysis of the collaboration network is crucial. Our results demonstrated that geographic location closeness and ethnic attributes has important roles in academic collaboration network establishment. Besides, it shows that famous scientific centers in the capital city of Iran, Tehran has strong influence on the production flow of scientific activities. These academic papers are mostly viewed and downloaded from the United State of America, China, India, and Iran. The motivation of this research is that by discovering hidden communities in the network and finding the structure of intuitions communications, we can identify each scientific center research potential separately and clear mutual scientific fields. Therefore, an efficient strategic program can be design, develop and test to keep scientific institutions in progress path and navigate their research goals into a straight useful roadmap to identify and fill the unknown gaps.

Due to extensive use of communication networks and the ease of communicating via wireless networks, these types of networks are increasingly considered. Usability in any environment without the need for monitoring and environmental engineering of these networks, have been caused increasing use of it in various fields. It also caused the emergence security problems in the sending and receiving of information that the intrusion detection has been raised as the most important issue. Hence, Network intrusion detection system (NIDS) is the process of identifying malicious activity in a network by analyzing the network traffic behavior. Wireless sensor network is a special type of wireless network is composed of sensors that are responsible for the task of collecting information from the environment. This wireless networks because of the limitation of resources, mobility and critical tasks are relatively high vulnerabilities in comparison to other network. There are several ways to secure a wireless network but those ways are not able to detect the majority of attacks. In addition, due to the limited power wireless sensor nodes, the use of observer nodes to permanent monitoring in wireless sensor networks in order to prevention and detection intrusion and attacks has practically impossible. Therefore, forecasting and intrusion detection systems play an important role in providing security in wireless sensor networks that can involve a wide range of attacks. Traffic behavior in the network has many features and dimensions, so dimensionality reduction plays a vital role in IDS, since detecting anomalies from high dimensional network traffic feature is time-consuming process. Feature selection influences the speed of the analysis and detection. For this purpose, in this project, a new approach is proposed to predict the intrusion of wireless networks using firefly based feature selection and fast learning network. Selected features in the feature selection phase are used as inputs to the fast learning network to analyze the intrusion of the network in real-time. According to the simulation results in this thesis it can be said that the fast neural network method continues training so as to avoid overfitting error. While neural networks further learns training set features until the training process is completed. Thus, the occurrence of overfitting phenomenon in neural networks is common. Therefore, the proposed method overall shows better performance than the neural network method in predicting new attacks on the network.

One of the problems of fog computing comparing to cloud computing’ s layers is the lack of hardware resources. Hence, this limitation causes restriction on the number of services that can be implemented simultaneously in fog layers and this restriction affects the services in dynamic networks. To illustrate, in fog networks that provide services to IOV, there are several different machines with contrasting services that each due to the movement of the IOV machines in local fog network, must be able to provide services to all local networks. Since there is not enough resources, this is not possible in the fog layers. The intention of this paper is providing a solution to optimize migration in fog computing. The solution for lack of the resource in fog layer that we present in this paper are: using a containerbased virtualization solution called LXC/LXD, and using CRIU for real time migration of container. Furthermore, to minimize the amount of transmitted data that takes most of the migration time, this paper will include the using of a lightweight template for LXC/LXD, called alpine which will reduce the amount of data migrated in the migration process.

Executing complicated computations in parallel increases the speed of computing and brings user delight to the system. Decomposing the program into several small programs and running multiple parallel processors are modeled by Directed Acyclic Graph. Scheduling nodes to execute this task graph is an important problem that will speed up computations. Since task scheduling in this graph belongs to NP-hard problems, various algorithms were developed for node scheduling to contribute to quality service delivery. The present study brought a heuristic algorithm named looking ahead sequencing algorithm (LASA) to cope with static scheduling in heterogeneous distributed computing systems with the intention of minimizing the schedule length of the user application. In the algorithm proposed here, looking ahead is considered as a criterion for prioritizing tasks. Also, a property called Emphasized Processor has been added to the algorithm to emphasize the task execution on a particular processor. The effectiveness of the algorithm was shown on few workflow type applications and the results of the algorithm implementation were compared with two more heuristic and meta-heuristic algorithms.

A common task in software maintenance is refactoring. The essential refactoring goal is improving software quality. The refactoring has extended for a wide range of software artifacts in different domains, hcus as web, database, and desktop applications. The difficulty of the manual artifacts inspecting is an impetus for automatic refactoring. Search-based software engineering )SBSE) has been used to solve many software engineering problems as optimization problems. Automating problem-solving is the goal of applying it to reduce human efforts. Many researchers have used the search-based approach in the field of refactoring that has known as search-based refactoring (SBR). There are many challenges in SBR. Identifying refactoring challenges help to understand refactoring problem and allow researchers to present new ideas for eliminating or reducing challenges. In this paper, we studied the researches in SBR and extracted researches challenges. We intend to categorize the challenges in this area from a new and more comprehensive perspective.

Relation extraction is the task of extracting semantic information from raw data. One of the key points in the area of open information extraction systems is the ability to extract relation information automatically for any domains, especially in web mining and web research. Many researches have been done in this field for relation extraction in different languages. Many relation extraction algorithms work based on parsing trees. The Persian language, as a low-resource language, has a dependency grammar and lexical structure which makes the dependency parsing difficult or timeconsuming, and it affects the speed of relation extraction in many cases. In this paper, we will introduce RePersian which is a fast method for relation extraction in Persian. Our proposed work is based on part-of-speech (POS) tags of a sentence and particular relation patterns. To achieve these patterns, we have analyzed sentence structures in the Persian language. RePersian searches through the POS-tags for finding the relation patterns, which are given in regular expression forms. In this way, RePersian finds semantic relations by matching the correct POS pattern to a relation pattern. We test and evaluate our method on the Dadegan, Persian dependency tree dataset, with two different POS tag-sets. Our approach had on average a precision of 78. 05% on finding the first argument of a relation, a precision of 80. 4% in finding the second argument and precision of 54. 85% on finding the right relation between the arguments.

Social media like Twitter have become very popular in recent decades. Hashtags are new kind of metadata which make non-structured tweets into searchable semistructured content. There are varied previous methods which recommend hashtags for new tweets. However, to the best of our knowledge, there is no previous word that uses BERT embedding for this purpose. In this paper, we propose a new method called EmHash that uses neural network based on BERT embedding to recommend new hashtags for each tweet. Unlike other word embeddings, BERT embedding constructs different vectors for the same word in different contexts. Emhash succeeded in outperforming three methods LDA, SVM, and TTM with respect to recall measure.

Today, the increasingly knowledgeable tourism system is one of the most diverse and largest industries in the world, with undeniable social and economic consequences and one of the most important sources of income for countries. Entering the digital age and the increasing growth of information and communication technologies, like all businesses and activities, has revolutionized the tourism industry. In the face of these rapid changes, "virtual tourism" is a common misconception that some non-tourism experts use as a type of tourism, and, surprisingly, even tourism students and graduates sometimes do so and they do phlegm. Due to the necessity of sound conceptualization that leads to better thought and practice, this study firstly explores the philosophy and concept of tourism through an analyticaldescriptive method and conducts library studies and then addresses the aspects of tourism development in the digital space, and it has shown that demand-driven virtual or electronic tourism is confused with demand-side virtual tourism. Since it is necessary to take into account the supply, demand and demand side of tourism, the use of the word virtual tourism is an obvious error and in conflict with the fundamental principles and philosophy of tourism. And the most important reason is that touring in cyberspace not only does not accompany detachment from daily life, which is a prerequisite of tourism, but in today's world, it is the same as the routineness of life.

Question Answering is a hot topic in artificial intelligence and has many real-world applications. This field aims at generating an answer to the user’ s question by analyzing a massive volume of text documents. Answer Selection is a significant part of a question answering system and attempts to extract the most relevant answers to the user’ s question from the candidate answers pool. Recently, researchers have attempted to resolve the answer selection task by using deep neural networks. They first employed the recurrent neural networks and then gradually migrated to convolutional neural networks. Nevertheless, the use of language models, which is implemented by deep neural networks, has recently been considered. In this research, the DistilBERT language model was employed as the language model. The outputs of the Question Analysis part and Expected Answer Extraction component are also applied with [CLS] token output as the final feature vector. This operation leads to improving the method performance. Several experiments are performed to evaluate the effectiveness of the proposed method, and the results are reported based on the MAP and MRR metrics. The results show that the MAP values of the proposed method improved by 0. 6%, and the MRR metric is improved by 0. 2%. The results of our research show that using a heavy language model does not guarantee a more reliable method for answer selection problem. It also shows that the use of particular words, such as Question Word and Expected Answer word, can improve the performance of the method.

Web services are nowadays providing users with a diverse range of services. They are mostly delivered through web applications. It is often the case that a single atomic web service cannot fulfill the demands of users. Rather, many simple atomic web services may have to be composed and form a complex one in order to handle users’ growing requests properly. Regarding the overall structure of web services as passive software components, they might fail to succeed properly, facing new types of requests. Recently, the concept of multiagent systems inspired many solutions in various research fields. In the web service composition domain, using smart agents so as to composite web services appropriately, leads to a complex, dynamic, and flexible service that meets different quality metrics altogether. In this study, a multiagent-based solution to web service composition is proposed using the “ TROPOS” methodology that handles incoming requests based on constructing task dependency graphs.

Today, due to the large volume of data and the high speed of its production, data analysis using traditional methods is infeasible. Data mining, as one of the most popular topics in the present century, has contributed to the advancement of science and technology in a significant number of fields. In recent years, researchers have used data mining extensively for data analysis. One of the important issues for researchers in this field is to identify common areas in the field of data mining and to find areas of active research in this field for future research. On the other hand, the analysis of social networks in recent years as an appropriate tool for examining the present and future relationships between the entities of a network structure has been used by scholars of various sciences to analyze and draw a scientific map of an area of science. In this paper, by using Co-Word Analysis and Social Network Analysis, the map and scientific structure of data mining topics in Iran based on articles indexed during the years 2009 to 2019 in the Civilica database are analyzed and the topic trends in the research field has been discussed. The results of the analysis show that in terms of data mining, concepts such as clustering, decision tree and neural network are the more important topics.

Epilepsy is a disorder in the electrical activity of the brain that occurs in a specific area or even the entire brain. These changes are visible through the acquisition of electroencephalogram (EEG) brain signals. EEG signals are important tools in predicting epilepsy because they are non-invasive measurement and display electrical activity at different external nodes at human brain. We used the CHB-MIT EEG Database in this study to develop an artificial model to predict epileptic seizures. Thus, we applied a one-dimensional convolutional neural network (CNN) to investigate raw EEG signals as an important indicator for starting time of a seizure. The seven-layer CNN was used to detect Preictal and Interictal states of brain where the performance of the proposed model was evaluated in terms of accuracy, specificity, and sensitivity which resulted in 97%, 98. 47%, and 98. 5%, respectively. Moreover, the proposed model was trained in two different feeding states: 1-Feeding by individual channel, 2-Feeding by grouped channels. It seems that the obtained results are promising.

Community Question-Answering websites, such as StackOverflow and Quora, expect users to follow specific guidelines in order to maintain content quality. These systems mainly rely on community reports for assessing contents, which has serious problems, such as the slow handling of violations, the loss of normal and experienced users' time, the low quality of some reports, and discouraging feedback to new users. Therefore, with the overall goal of providing solutions for automating moderation actions in Q&A websites, we aim to provide a model to predict 20 quality or subjective aspects of questions in QA websites. To this end, we used data gathered by the CrowdSource team at Google Research in 2019 and fine-tuned pre-trained BERT model on our problem. Based on our evaluation, model achieved value of 0. 046 for Mean-Squared-Error (MSE) after 2 epochs of training, which did not improve substantially in the next ones. Results confirm that by simple fine-tuning, we can achieve accurate models in little time and on less amount of data.

Online medical reviews contain patients’ subjective evaluations and reflect their satisfaction with the treatment process and doctors. Mining and analysis of sentiment expressed in these medical data may be vital for different applications including adverse drug effects detection, doctor recommendation, and healthcare quality assessment. Nevertheless, medical sentiment analysis is a challenging and complex task because patients who write the reviews are usually non-professional users and tend to use informal language. The problem is more challenging in the Persian language due to its resource scarcity and complex structure. In this study, we introduce PODOR, a Persian dataset of online doctor reviews extracted from social web. Also, we propose a deep model based on the bidirectional long short-term memory for polarity detection of PODOR reviews. To show the effectiveness and suitability of the proposed model, we compared the model with six traditional supervised machine learning methods and three deep models. Preliminary comparative results indicated that our model outperformed traditional methods by 8% and 7%, and deep models by 2% and 3% in terms of accuracy and f1-measure.

Internet of things and its applications have been growing rapidly in recent years. IoT is included in various communication networks and technologies that part of those are sensor networks. Due to battery limitations, these networks must be designed to reduce power consumption and extend network life. Clustering-based methods save energy by balancing the energy consumption of the nodes. Cluster size has a huge impact on saving energy and increasing network lifetime. This paper presents a new virtual network-based clustering method using multi-phase intelligent algorithms with respect to the energy of the nodes. The proposed method can increase the lifetime of the network and decrease the energy consumption of the nodes by accurately identifying the number of suitable clusters, distributing the cluster-head in the environment, increasing time of the phase of steady, and greedy routing. The proposed method has been evaluated after simulation in Matlab software and its results have been compared with other algorithms. The results of the evaluation show that the proposed method improves the energy consumption and network lifetime.

Instagram is one of the popular social media services used by a variety of people around the world. It has a huge number of active users. The more users, the larger and the more different Instagram data are available. In this paper, we propose a Model-based location recommender system (MLRS), which creates a profile for each location and uses it to recommend locations, based on user interests. Since our analysis does not have an appropriate dataset to check, we use both Foursquare and Instagram to create our dataset. Next, we propose the Term-Frequency and Inverse Document Frequency(TF-IDF) method to rank extracted hashtags of selected Instagram locations based on Instagram image captions. This gives us the main idea of locations, based on 30 recent image captions hashtag posted. Then, we used FastText to classify hashtags of each location post. We evaluated our system with a large-scale real dataset collected from Instagram concerning precision, recall and the F-measure. Finally, the experimental results show that the highest result achieved when the FastText model tested with n=1 with an F-measure of 77. 8%.

The tremendous increase in Web usage led to the appearance of different network structures. One of the essential issues in the field of network science and engineering is to find and utilize network structures such as community structures by community detection. Although most of the current algorithms for detection of community use on the binary representation of the networks, some networks can encode more information instead of the topological structure, in which this information can be applied appropriately in detecting communities. Network information can be represented in the form of weights and identified as the weighted social network. This paper proposes a new algorithm using irregular CLA (cellular learning automaton) for finding the community in weighted networks called CLA-WCD. The CLA-WCD can find near-optimal community structures with reasonable running-time by taking advantage of the parallel capability and learning ability of the cellular automata and learning automaton, respectively. The CLA-WCD is also evaluated on real and synthetic networks in comparison with popular community discovery methods. The simulation results demonstrated that the CLA-WCD outperforms other methods.

Influence maximization is a problem based on diffusion and probability in social networks with the aim of finding the least node with the most influence. These nodes play an essential role in the diffusion process. However, the influence maximization problem faces two essential challenges of time efficiency and optimal selection of the seed nodes. To solve these challenges, we proposed an algorithm based on the properties of the graph topology structure and centrality, called IMT (Influence Maximization based on the Topology) algorithm. This algorithm selects the seed nodes from the dense part of the graph that can access more nodes in the shortest distance. Finally, experiments showed that the proposed algorithm outperformed the other algorithms in terms of influence spread and running time.

One of the emerging technologies that have entered the daily life of human beings with the development of wireless networks is the Internet of Things (IoTs). IoTs is the invention of new intelligent devices, which is rapidly developing. In this paper, we implement the application layer protocols and two of the most trusted IoT protocols, MQTT and CoAP in terms of power consumption and implement them on Nodemcu hardware. The results in the graphs show that the average power consumption of the MQTT protocol is higher than the CoAP protocol, which is due to the high reliability of the MQTT protocol, which imposes additional overhead on the network, resulting in additional overhead. It is a control flow mechanism that is used in the TCP sub-layer protocol as the MQTT network transmission layer protocol.

Following the dramatic growth of malware and the essential role of computer systems in our daily lives, the security of computer systems and the existence of malware detection systems become critical. In recent years, many machine learning methods have been used to learn the behavioral or structural patterns of malware. Because of their high generalization capability, they have achieved great success in detecting malware. In this paper, to identify malware programs, features extracted based on the header and PE file structure are used to train several machine learning models. The proposed method identifies malware programs with 95. 59% accuracy using only nine features, the values of which have a significant difference between malware and benign files. Due to the high speed of the proposed model in feature extraction and the low number of extracted features, which lead to faster model training, the proposed method can be used in real-time malware detection systems.

This paper discusses the role of social media in shaping the life of an individual who publicly challenges Islamic authoritative discourse (i. e the hadith of Prophet Muhammad and the Quran). Sally (pseudonym), a 49 year old Malaysian lady, was the sole participant of this study where the impacts of social media on her life were investigated indepth. Data were generated using an ethnographic approach involving a prolonged observation of her social media accounts (i. e Facebook, Blog and Twitter) for four years from 2014 to 2017. This time period covers the major developments in her life since her public rejection of Islamic authoritative discourse to her settlement in the USA. A thematic analysis was carried out to make sense of the data. The findings reveal that social media has affected her life in both negative and positive ways. She faced several problematic experiences (e. g. break up with family, life threats, police reports and exile) due to her postings which challenge Islamic authoritative discourses, however she also made full use of the social media to finally put a stop to the problematic experiences.

The Internet of Things (IoT) is a new concept in the world of technology and communications for the description of the future in which physical objects or sensor nodes connect to the Internet using IPv6. One of the most prominent protocols used on the IoT for routing is RPL (IPv6 Routing Protocol for Low power and Lossy Networks) that could be exposed to specific attacks like the Local Repair attack. Hence, researchers focus on RPL security as the most important challenge at this protocol. In this paper, we proposed a new Fuzzy-based method for the detection of Local Repair Attack on the RPL routing protocol. The obtained results using the Cooja simulator in the Contiki OS show that the proposed method detects the Local Repair attack with a very high True Positive Rate (TPR) and very low False Positive Rate (FPR).

One of the key issues in Vehicular Ad-hoc networks (VANETs) is to optimize the traffic congestion. Cooperation in these networks is a challenging issue due to their specific characteristics. In this paper, a non-cooperative game theory-based approach is introduced for packet forwarding. Through extensive mathematical analyses and also experimental validation, we prove that the proposed non-cooperative game mechanism attains the Nash equilibrium point. Our designed mechanism encourages all vehicles to collaborate with each other in packet forwarding operations. This, in turn, results in decreasing the payments by nodes to the network side and also results in optimizing traffic congestion. The simulation results established the robustness of the proposed mechanism in terms of cost-related criteria.

Virtual reality is novelty technology with many applications in various fields. The high cost of interactive virtual reality hardware is one of the challenges that can affect on widespread applications. However, it can be facilitated with the development of affordable devices. In the present study, three low-cost accurate modules were designed for the measurement of orientations and sent them via Wi-Fi in the virtual reality environment to motion transfer. ADXL 345, MPU 6050 and MPU 9250 were used to build modules. Kalman and complementary filters were applied to the reduction of noise and drift, as also the digital motion processor of MPU 6050 and MPU 9250 with the quaternion base method was used in the second method. The built-in modules were obtained good precision, however, the accuracy of the pitch and roll axes were better than the Yaw. ADXL 345, and MPU 6050 modules (Inertial Measurement Units) are affordable for 2D and 3D orientation measurements, respectively. Our products can be applied in any field that needs simulation motion in a virtual reality environment.

Recently, facility on the internet access in worldwide has caused many businesses take their activities on the internet affiliate networks. But security threats, such as phishing attacks, have always threaten these businesses. The multiplicity of web pages features has led to use of feature selection methods and their combination with machine learning methods to detect phishing. In this paper, Penguin metaheuristic algorithm and its performance are investigated to find the optimal response to phishing detection as the main contribution. Therefore, we propose a combination of Penguin algorithm in feature selection phase with artificial neural network in the phishing detection phase. Also, in order to train and evaluate our proposed method, a dataset with 11055 samples of phishing and normal websites is used. The results of our proposed method using the implementation in MATLAB software present that with increasing the population size and the number of iterations in penguin optimization algorithm, the average value of the feature selection function decreased by 69. 57% and the RMSE index reduced by 24. 56%. Finally, our proposed method shows about 29. 16% lower error in phishing detection in comparison to multilayer artificial neural network.

Mankind's present life has progressed in such a way that objects play a vital role. In the past, objects functioned as a separate entity in a limited domain. But modern IoT technology, where communication goes beyond human-machine to machine-to-machine communication, has been assumed to enable data functions in objects and to see semi-intelligent functions in objects. IoT is a new telecommunications paradigm that envisions a near future that, by connecting virtual and physical objects to the Internet, will automatically send and interact with the environment, making these intelligent objects an integral part of the Internet. IoT can also be a good solution for green information technology because it has high potential, is easily updated and renewable, and can have significant environmental benefits. On the other hand, in the last few years, the topic of green libraries, which has made them ideal locations for the emergence of green technology and strategically become a sustainable model for society, has received much attention from researchers. In this study, while reviewing the concept of IoT in general, the use of IoT in green libraries is presented to provide better services and energy saving. In this regard, we will explain some of the IoT technologies such as Green RFID, Green Wireless Sensor Network, Machine-to-Machine Communication and Green Data Centers, and finally the principles and rules of IoT technologies and recommendations. It is referred to for future research.

Text Simplification is described as the process of transforming natural language text, both lexical and syntactic. The structure and grammar of the output must be considerably simplified, and understandability and readability should be improved, while original meaning and information are maintained. Text simplification is a fast-growing domain and can be used for several applications, such as pre-processing tool in the pipeline of natural language processing tasks as well as helping people with special needs. While the automated text simplification operation itself is challenging, the bench-marking and evaluating of this task is even more elaborate and controversial. Although there have been some reviews in the context of text simplification, no specific survey has been done on the question of evaluation metrics and methods of text simplification. In this paper, we review the most significant studies identified out of more than 300 studies of the last three decades in the field of text simplification, focusing on evaluation metrics, methods and corpora. There are different evaluation metrics, methods and corpora for text simplification based on approaches, datasets, and algorithms used to perform simplification task. We made a review of these different metrics and provided results of high-quality research studies on each criterion.

Steganography is the task of hiding information in some media normally images. Steganalysis is the process of discriminating such instances and clean ones. In recent years, steganalysis has tended to use deep learning for feature extraction and classification. Convolutional Neural Networks (CNN) have improved the steganalysis performance but at the cost of computational complexity and memory space due to huge amount of training data. In this paper, a new framework is proposed to reduce the learning cost by a divide and conquer strategy. In the first phase, data is divided into disjoint clusters by use of k-means. Each cluster is then fed to a separate CNN to be customized on a specific region of data space. In the final phase, the networks are merged leveraging a fast alternate-weighting process. The proposed weighting can, to some extent, compensate for reducing the size of training data per model. The experimental results show that the proposed scalable framework reduces memory and time complexity with preserving accuracy.

Increasing and ease of access to big data have led to new developments in e-learning and increased attention and research in the field of learning analytics. The purpose of this research is to introduce The Social Network Analysis (SNA) method as one of the new applications of learning analytics for detecting learning networks in e-learning environments. This study discusses the importance of social network analysis and its definition, then introduces its applications in different domains, followed by analytical methods and evaluation criteria in social network analysis. The research results show that because of the achievements of multi-disciplines like mathematics, sociology, computer science and education, social network analysis is an interdisciplinary and high-performance domain for studying and demonstrating communication patterns that have never been easier to use. For this reason, it requires communities of professionals with different disciplines.

By the growing number of viruses and also epidemics, predicting and controlling the epidemics have high priority in today's human life. Network theory is a useful instrument for modelling the epidemics. As we can see, some predictions have been proposed for the disease like influenza (N1H1 virus). In this paper we aimed to compare the spreading model of coronavirus with proposed epidemic models. Also, we have shown that informing people using impressive ways such as social networks and also preventing attempts done by the governments affects the transmission rate. So models which are formed based on static transmission rate are not applicable for disease with dynamic transmission rate.

With the rapid growth and development of cloud computing, many startups or even large enterprises have decided to employ cloud services due to its dynamic provisioning and also reducing costs through the reduction of seasoned human resources needed to maintain the infrastructure. On the other hand, the risks of using shared resources and virtual machines are among the biggest threats that have always challenged the advantages of using cloud services. It is now possible for an attacker to attack a virtual machine and through that target the virtual machine manager or hypervisor. This approach will allow the attacker to completely control other virtual machines existing in the domain. Therefore, using a mechanism to decrease the colocation degree of vulnerable virtual machines on the same physical machine can be effective in reducing such security risks. To address these issues, we have proposed a security-aware virtual machine placement scheme to reduce the risk of vulnerable virtual machines co-placement. To manage the precision of security evaluation, it is vital to consider some hesitancy factors regarding security evaluations. Thus, to consider hesitancy in our proposed method, hesitant fuzzy sets are applied and several experimental evaluations have demonstrated the benefits of the proposed approach.

Given the growing social media presence, the content available on social media sites can have a much wider impact on users than traditional marketing. For this reason, the purpose of this research is to identify the visual features of posts on the Travel Agency Instagram page and to discover the interest of followers from different countries in modeling to improve and develop social media marketing. This research is an applied research in terms of purpose and is a descriptive survey in terms of data collection and analysis. The information used for conducting the research was obtained using systematic database scrolling and programming. Data analysis was performed on three groups of images with different themes and constant color spectrum. Research process based on standard CRISP-DM model and decision tree with CHAID algorithm is also used for modeling. The results show that the subjective nature of the posts has a positive effect on the interest or dislike of followers in different countries And the countries of Egypt and the UAE are among the valuable countries to allocate marketing and investment funds to enhance the Agency's profitability.

with the rapid growth of using the internet and social media, people can easily share their opinions on these platforms. due to this fact, user's comments are considering as a rich source for natural language processing tasks such as sentiment analysis. This paper represents our work on producing Insta-text, which is an Instagram comments Persian language sentiment analysis dataset. In this study, about 111, 000 Instagram comments have been scrapped and about 9, 000 of them have been labeled using the crowdsourcing method. Word2vec model also has been used to validate the dataset.

Recently, social networks sites have become a popular phenomenon. The growth of these sites is rapid and exponential. Also, Social Network Sites are attracting researchers interested in online technologies and user behavior. These research include various aspects of social network sites such as, users’ motivations and attitudes toward social networking sites. But despite the importance of user satisfaction, there is a few research on this subject in social network context. Given the growth of these sites in Asian countries, and in particular Iran, studying them is very important. This study uses thematic analysis to examine Determinants of social network’ s user satisfaction. The results of this study could lead to improved interaction with social network users. The results revealed seven key themes: time responsiveness, utilitarian values, hedonic values, ease to use, format, perceived privacy and personalization.

Opinion mining and sentiment mining as a sub fields of data mining and machine learning nowadays are interesting research fields due to the increasing number of user roles and their attempts in declaring their opinions in various domains of social networks and e-commerce. These areas are closely related to artificial intelligence and especially natural language proceesing. For this reseaon, employing developed algorithm with high accuracy for other languages in another language especilaay Persian is very difficult and sometimes impossible. The number of conducted research on Persian language is a few ones and deploying the results of such algorithms on e-commerce domain is necessary and of high importance. There are some associated challenges for conduction such research such as the the lack of graph based dictionarires including the sentiment words, the graph based dictionaries including the dependency of words and the degree of their dependency and softwares for determining the sentence's part opf speech. Some of such tools are existed but either are incomplete or they are not limited to a special domain which make using them very challenging and time consuming. This paper describes our attempt to construct a graph based dictionary in Persian language including the opinion and sentiment words of different part of speech for hoteling and tourism domain that using them for researchers leards to higher accuracy and faster processing time in the related areas.

Clustering is one of the fundamental techniques of data mining that is used for dataset analysis. Clustering algorithms group available data based on similarity or distance measures. Two important clustering methods used in the literature are hierarchical and density based methods. A lot of algorithms have been developed based on these two concepts separately. Birch and its extensions are samples of hierarchical based methods. DBSCAN and its extensions are samples of density based methods. In this paper, a new algorithm is proposed to use both concepts together to achieve an acceptable speed and results, simultaneously. At first, it tries to make clusters using a hierarchical method. If it decides to make a new cluster, then the algorithm checks for density. In this manner, it tries to postpone splitting the clusters. To show the effect of the proposed algorithm, some evaluations are performed on some synthetic and real datasets which show some improvements over related works.

Influence Maximization is one of the most popular problems in social networks, which has many applications in real-world networks. Influence maximization is the problem of finding a small subset of nodes (seed nodes) in a social network that could maximize the spread of influence. However, fairness has not been considered widely in this domain. An important question is whether the benefits of such information propagation in a social network is fairly distributed across different groups in the population. Thus, single-objective influence maximization problem turns to the multi-objective problem, in which both the spread of influence and fairness must be considered. In this paper, we first introduce the measure we use for measuring fairness in influence maximization. Then, we propose an algorithmic framework, based on multi-objective grey-wolf optimizer, for multi-objective optimization of combinatorial problem, and use it to solve the explained problem. Finally, we compare our results on real-world datasets with some greedy algorithm to show the effectiveness of the proposed method. The results show that our algorithm improves both influence and fairness in the network.

The e-learning environment includes various types of learners who should take responsibility for their learning. Personalization of the e-learning environment is a significant contributing factor in the effective learning process which enhances learning satisfaction, speed learning, quality and efficiency of the learning process. The major goal of personalization in e-learning is providing appropriate education as well as adjusting the environmental conditions for each learner according to their specific characteristics. Various factors could be considered for designing a personalized learning environment; this study focused on the selective learning style. In this research, the e-learning environment was personalized based on Kolb’ s learning style and tailored learning strategies were developed for learners. Finally, in the designed learning environment, the learners' performance in an e-learning course for 19 students was evaluated. The results showed that personalization in the e-learning environment based on learning style influences students' academic success and satisfaction significantly.

Community question answering is a forum in which users ask their questions and other existing users respond those questions. As these forums grow, their users, questions and answers increase significantly and new challenges appear. For instance, someone who has asked a question in forum has to wait a lot to another user give him a response. On the other, experts has to spend a lot of time finding questions in their expertise field. Thus, the most important challenge which these forums face is finding appropriate users to answer the questions. One of the most critical issues which expert finding algorithms have is vocabulary gap between skills and words used in posts text. Therefore, in this paper, we propose two translation models to reduce vocabulary gap. The proposed models are evaluated using the StackOverflow dataset and compared with the related algorithms. The experimental results show the superiority of the proposed models.

Cardiovascular disease is one of the most common causes of mortality in the world. Among the different types of this disease, the coronary artery is the most important, which the correct and timely diagnosis of which is vital. Diagnostic and treatment methods of this disease have many side effects and costs. The best and most accurate diagnostic method here is angiography. Researchers seek to find economical and highaccuracy methods for this purpose. The disease-related features and different data mining techniques are described to increase the accuracy of the diagnosis through one dataset of essential and useful features. Data are collected from 303 suspected cardiovascular patients in Shahid Rajaee Hospital, Tehran. Among the samples, 87 are healthy, and 216 are sick. The features are selected through their optimal subsets of performance, speed of diagnosis, and precision in the first step to determine the severity of coronary artery disease (CAD). This feature selection can predict and promote a learning model. Then the optimal machine learning models are applied to analyze and predict CAD. The accuracy of 99. 67% is found in this diagnosis, indicating the highest obtained accuracy in this field. The left anterior descending (LAD), the left circumflex (LCX), and the right coronary artery (RCA) features are diagnosed with high accuracy by using those models. It seems these three features define the CAD and are dependent on angiography. If they are eliminated for the prediagnosis situation, the accuracy of CAD will be between 83% to 86% for the new reduced subset of features proposed concerning legible performance reduction.

Sport is regarded as an inseparable part of human life. Currently, a growing trend is observed in people's interest in football teams. In general, a successful procedure in players’ communication is one of the main factors required for the victory of that team. The present study aimed to perform analyzes based on the perspective of social and communication networks (such as player passes and in-game transactions) to improve team performance. The analysis was performed on data collected from three matches of the Persepolis club in the first half-season of the Iranian Premier League 2019-20. This research seeks to review this issue from two integral perspectives as follows: 1) evaluating the performance of individuals as a part of a social network, 2) investigating the communication network between players. To this aim, we used the innovative method of recognizing and classifying frequent subgroups in this analysis. It is worth noting that 20 persen of these routes were in the defensive line while 31 persen were in the defensive midfielder. However, there were no routes in the attacker line or offensive midfielder, which indicated a form of weakness. On the other hand, various types of node degrees, points, and n-pass cycles were calculated in other sections. The results revealed the weak performance of the connection bridge between the team's playmakers and the end-players for shooting the ball. Although these topics were discussed at a minor level and only three matches of a team, the results can be generalized to other issues.

Considering the growing trend of electric power consumption, resource constraints and the exhaustion of existing grid equipment, the issue of restructuring the electricity industry has been considered. Meanwhile, the use of Internet of Things (IoT) technology and upgrading the power grid to a Smart Grid (SG), in addition to the many benefits, poses challenges to security issues. Since Intrusion Detection System (IDS) is one of the ways forward to combat cyber-attacks, Therefore, in this paper, a smart method for intrusion detection in these types of networks is presented. In this method, a combination of three decision trees was used to detect intrusion and the performance of the proposed method was compared with the Support Vector Machine (SVM), K-Nearest Neighbors (KNN) and Decision Tree (DT) methods. Experiments have been performed on the NSL-KDD dataset and the results show that the proposed method performs better than other methods for Intrusion Detection in IoT-Based SG.

Smart accident detection systems are systems that automatically detect accidents. IoT (Internet of Things) can connect different hardware, especially sensors, and can do processes using the information received from sensors, which lead to accident detection; and then this information is made available to relevant authorities such as medical emergencies and the police. In this research, some of the most recent methods for car accident detection using IoT technology are investigated and then obstacles, challenges and future trends are analyzed and compared.

Nowadays, there are many software repositories, especially on the web, which have many challenges to be automated. Duplicate bug report detection (DBRD) is an excellent problem of software triage systems like Bugzilla since 2004 as an essential online software repository. There are two main approaches for automatic DBRD, including information retrieval (IR)-based and machine learning (ML)-based. Many related works are using both approaches, but it is not clear which one is more useful and has better performance. This study focuses on introducing a methodology for comparing the validation performance of both approaches in a particular condition. The Android dataset is used for evaluation, and about 2 million pairs of bug reports are analyzed for 59 bug reports, which were duplicate. The results show that the ML-based approach has better validation performance, incredibly about 40%. Besides, the ML-based approach has a more reliable criterion for evaluation like accuracy, precision, and recall versus an IR-based approach, which has just mean average precision (MAP) or rank metrics.

Since most of the energy consumption is related to buildings, energy management in smart homes is a major challenge. Personalized recommender systems are a solution to optimize energy consumption by analyzing building energy consumption behaviors. The NILM energy disaggregation technique has been considered in recent years. However, the combination of recommender systems and NILM has received less attention. This paper proposes a personalized NILM-based recommender system that has three main phases: DAE-based NILM, TF-IDF-based text classification, and personalized recommendation. Because of the noise in the energy data, the DAE-based NILM helps detect these noises from the signals. Households’ requirements and interests are identified at this stage. In the second phase, the TF-IDF technique is used to extract meaningful keywords from the advertised optimal tags and assign them a label. Finally, in the third phase, the cosine similarity technique is used to provide some recommendations. This step generates a suggestion for each device that is on the requirement list. The proposed approach was tested using the REDD dataset. The results showed that the accuracy of the recommendation system was about 60%.

Data clustering aims to discover the underlying structure of data. it has many applications in data analysis and it is one of the most widely used tools in data mining. DBSCAN is one of the most famous clustering algorithms. its advantages are to identify clusters of various shapes and define the number of clusters. Since DBSCAN is sensitive to its parameters which are ε and MinPts, it may perform poorly when the dataset is unbalanced. To solve this problem, this paper proposes a sliding window DBSCAN clustering algorithm that uses Gridding and local parameters for unbalanced data which we will refer to as SW-DBSCAN. The algorithm divides the dataset into several grids. The size and shape of each gird depends on the specimen density specification. Then, for each grid, the parameters are adjusted for local clustering and eventually merging data zones. Experimental results show that this algorithm can help to improve the performance of the DBSCAN algorithm and can deal with arbitrary data and asymmetric data.

In recent years, data mining has been widely used in the analysis of educational processes and the analysis of learners' behaviors in MOOCs. Therefore, in this study, we investigate different methods for analyzing learners' behavior in MOOCs as well as investigating factors of satisfaction and dissatisfaction and providing appropriate solutions for them. Through this study the most important methods used for data analysis are classification and clustering methods. Also the most important factors that make the students satisfied and dissatisfied with these courses are the reputation of the universities, the quality of the courses, the flexibility of the course, the time bruise, the lack of motivation, the isolation and the lack of interaction in the course and sufficient scientific background. The results of this study can help to gain a holistic view of the various elements of effective MOOCs and increase the rate of completion of the course.

The strength of information diffusion on social networks depends on many factors, including the selected influential nodes. The problem of finding such nodes in the network is modeled by influence maximization problem, which faces two essential challenges: (1) inadequate selection of the seed nodes due to the lack of focus on the rich-club phenomenon and (2) high running time due to the lack of focus on pruning the graph nodes and localization. To solve these challenges, a computational localization-based RLIM algorithm is presented here to prevent the rich-club phenomenon. In this algorithm, the graph nodes are pruned based on the eigenvector centrality to reduce the computational overhead, and then the computations are performed locally using localization criteria. After that, influential nodes are selected by avoiding the rich-club phenomenon. In the RLIM algorithm, the seed nodes provided a better influence spread than the other algorithms. Experimental results on the synthetic and real-world datasets shows that the RLIM algorithm can verify the high effectiveness and efficiency than the comparable algorithms for an influence maximization problem.

The dark web has the potential to host an increasingly high number of legal and illegal activities ranging from maintaining privacy to selling illegal goods, mainly purchased with Bitcoin or other digital currencies. They may be used to circumvent censorship, access blocked content, or maintain the privacy of sensitive communications or business plans. However, a range of malicious actors, from criminals and terrorists to state-sponsored spies, can also leverage cyberspace and the Dark Web can serve as a forum for conversation, coordination and action. It can also be used by governments to shield military personnel and Social change in other countries. It has remained largely unregulated by the government, and the first step in better monitoring and policing the Dark Web is better understanding it. The topic of describing general atmosphere of the dark web from people's activities to the presence of governments, providing a proper understanding of the dark web, In order to establish effective policies and raising awareness about the formation, onion router working, concept of anonymity, hidden services, active groups and ultimately the presence of governments.

Routing is one of the major challenges in designingwireless body area networks. Evolutionary-based Multihop Routing Protocol for wireless body area network is the first protocol offered for body surface networks. Some disadvantages of this protocol are as 1-It uses a genetic algorithm to adjust program parameters, which has the disadvantage that it performs differently in two runs, has poor mathematical support, requires a lot of computation, a good answer But it may not find the optimal answer. 2-Although the knots are behind the body, a sink is just in front of the body, so the knots in the body have to go a long way and spend more energy to communicate with the sink. To overcome these drawbacks, a two-sink based Particle Swarm Optimization algorithm is proposed for wireless body area networks. Instead of the genetic algorithm, the particle swarm optimization algorithm is used, which has a simpler concept, faster convergence and easier implementation than the genetic algorithm and has a smaller population size. It also uses another sink behind the body, which makes most dorsal nodes use single step communication to reduce path loss and energy consumption.

The software Defined Networks, by separating the data plane and control plane of the network, have made a drastic change to the scope of computer networks. Although this separation has accelerated and simplified the management, configuration and error detection, it has also caused some new security problems. One of these problems is the Vulnerability of the software defined networks’ architecture to distributed denial of service attacks on the network’ s controllers. One of the most recent distributed denial of service attacks which entropy-based methods are incapable of detecting, is to send fake packets with different source to random addresses in a software defined network. In this paper, given the SDN structure and traff ic analysis, a statistical trapezoid model is introduced to estimate number of table misses for each switch. Then, using the linear regression method and EWMA estimation, the threshold of the table misses in specified time intervals, is estimated. The evaluation results imply that using this method, one can detect DDoS attacks in early stage in software defined networks, regardless of the sort of DDoS attack.

Trust network is widely considered to be one of the most important aspects of social networks. It has many applications in the field of recommender systems and opinion formation. Few researchers have addressed the problem of trust/distrust prediction and, it has not yet been established whether the similarity measures can do trust prediction. The present paper aims to validate that similar users have related trust relationships. To predict trust relations between two users, the LookLike algorithm was introduced. Then we used the LookLike algorithm results as new features for supervised classifiers to predict the trust/distrust label. We chose a list of similarity measures to examined our claim on four real-world trust network datasets. The results demonstrated that there is a strong correlation between users' similarity and their opinion on trust networks. Due to the tight relation between trust prediction and truth discovery, we believe that our similarity-based algorithm could be a promising solution in their challenging domains.

Human action recognition in video is one of the most widely applied topics in the field of image and video processing, with many applications in surveillance (security, sports, etc. ), activity detection, video-content-based monitoring, man-machine interaction, and health/disability care. Action recognition is a complex process that faces several challenges such as occlusion, camera movement, viewpoint move, background clutter, and brightness variation. In this study, we propose a novel human action recognition method using convolutional neural networks (CNN) and deep bidirectional LSTM (DB-LSTM) networks, using only raw video frames. First, deep features are extracted from video frames using a pre-trained CNN architecture called ResNet152. The sequential information of the frames is then learned using the DB-LSTM network, where multiple layers are stacked together in both forward and backward passes of DB-LSTM, to increase depth. The evaluation results of the proposed method using PyTorch, compared to the state-of-theart methods, show a considerable increase in the efficiency of action recognition on the UCF 101 dataset, reaching 95% recognition accuracy. The choice of the CNN architecture, proper tuning of input parameters, and techniques such as data augmentation contribute to the accuracy boost in this study.

The primary goal of cognitive computing is to design a digitalized model that is able to mimic human thinking processes. The cognitive engine is in charge of implementing the functionality of a cognitive system. Nowadays, cognitive engines are used as a self-organized management mechanism in different fields such as computer networks, Internet of Things (IoT), and Robotics. This is because the management algorithms of these fields are going to be very complex and therefore human thinking models as digitalized models are required for fast and accurate decision making. In this paper, we summarize challenges in designing cognitive engines. Then, a set of challenges in designing the cognitive engine for body-mind operating system in the digitalized healthcare system is obtained. In the literature, our survey and also suggested case study in the healthcare system have not been considered yet.

This article provides a comprehensive literature review of blockchain protocols for IoT. First, we describe the blockchain and summarize the research involved with blockchain technologies. Then, we provide an overview of the application areas of blockchain technologies in IoT, for example, the Internet of Vehicles, Energy Internet, Cloud Internet, Edge Computing and more In addition, we provide a categorization of threat models that are divided into five categories by blockchain protocols in IoT networks, namely: identity-based attacks, manipulation-based attacks, coding attacks, Credit-based and service-based attacks. In addition, we provide a peer-to-peer classification and comparison of modern methods for securing and protecting the privacy of blockchain technologies with respect to blockchain model, specific security objectives, performance, constraints, computational complexity, communications. Based on current research, we highlight emerging research challenges and discuss possible future research directions in blockchain technologies for IoT.

Machine Translation (MT) refers to the automated software-based translation of natural language text. The embedded complexities and incompatibilities of natural languages have made MT a daunting task facing numerous challenges, especially when it is to be compared to a manual translation. With the emergence of deep-learning AI approaches, the Neural Machine Translation (NMT) has pushed MT results closer to human expectations. One of the newest deep learning approaches is the sequence-to-sequence approach based on Recurrent Neural Networks (RNN), complex convolutions, and transformers, and employing encoders/decoder pairs. In this study, an attention-based deep learning architecture is proposed for MT, with all layers focused exclusively on multi-head attention and based on a transformer that includes multi-layer encoders/decoders. The main contributions of the proposed model lie in the weighted combination of layers’ primary input and output of the previous layers, feeding into the next layer. This mechanism results in a more accurate transformation compared to nonhybrid inputs. The model is evaluated using two datasets for German/English translation, the WMT'14 dataset for training, and the newstest'2012 dataset for testing. The experiments are run on GPU-equipped Google Colab instances and the results show an accuracy of 36. 7 BLEU, a 5% improvement over the previous work without the hybrid-input technique.

Robustness is one of the most important properties of network, because it represents the network tolerance against failures. Thus, this property is considered in many real-world networks, such as distribution networks and communication networks. In order to evaluate this property, many measurements have been proposed, most of which are based on the size of giant component. In this paper, we introduced a concept of coloring nodes, by which we can classify nodes into two groups, and based on this concept, we proposed a new metric to measure the network robustness. Then, we implemented our proposed metric on random network and scale-free network to compare their behaviors. Finally, we compare the efficiency of our proposed method with another state-of-the-are robustness metric.

Logs contain valuable information about user actions on the web that are widely used in a variety of security, industry and science areas. Logs are an excellent resource for determining the health of a system. An enormous amount of logs are generated every day that record users' activities, which is very difficult to analyze in traditional ways, so there is a need for intelligent and cognitive analytics. Machine learning techniques are currently used as an efficient tool in the processing and analysis of web logs. In this paper, we used a 5-layer neural network with 2461 parameters and sigmoid, hyperbolic and relay activation functions, in addition to intelligently detecting cyber threats to optimal results in accuracy and minimizing error function. In analyzing web logs compared to other conventional machine learning methods such as clustering, decision tree, support vector machines, principal component analysis, isolated forest and logistic regression.

Being aware of people's attitudes and emotions of a person or event can have a huge impact on the decisions of individuals and organizations. With the rise of social networks and in particular Instagram because of its popularity, many people are sharing their attitudes on this network. Analyzing the emotions of users of this social network, which is a great example of community, can help managers make organizational decisions and predict important events such as elections. In this research, the EAS system was designed and implemented to extract emotion and visualize them. As a practical example, the Instagram users' feelings about the two main candidates for the 12th Iranian presidential election are also examined. The result shows a positive feeling about the word " روحانی " and a high level of confidence emotion along with anger and disgust emotions.

The growth of information and communications technology and consequently need to secure data transfer has made steganography as an attractive and essential issue in recent years. Steganography techniques try to hide the original message in other forms of data, the most common of them are digital images. Although the methods based on Least Significant Bit (i. e. LSB) are widely used as the basis for many steganography algorithms in spatial domain, but some challenges such as limited capacity and efficiency may hamper the usability of these methods. This paper presents a modified version of the LSB based methods which hides the original message by utilizing the concept of transferring to even quantities in host image. Implementation and testing of the proposed method shows that this scheme may significantly increase the signal-to-noise ratio of the hiding procedure as well as the structural similarity compared to alternative methods.

Traffic classification plays an important role in network management and cyber-security. With the development of the Internet, online applications and in the following encrypted techniques, encrypted traffic has changed to a major challenge for traffic classification. In fact, unbalanced data, in which the unbalanced distribution of samples across classes lead to the classification performance reduction, is considered as one of the prominent challenges in encrypted traffic classification. Although previous studies tried to deal with the class imbalance problem in the pre-processing step using machine learning and particularly deep learning models, they are still confronting with some limitations. In this regard, a new classification method is proposed in this paper that tries to deal with the problem of unbalanced data during the training process. The proposed method employs a cost-sensitive convolution neural network and considers a cost for each classification according to the distribution of classes. These costs are then applied to the network along the training process to enhance the overall accuracy. Based on the empirical results, the proposed model obtained higher classification performance (about 2% on average) compared to the Deep Packet method.

Considering the vital issue of disaster management and the requirement of its additional development due to the increase in the rate of disasters over the recent years, using new technologies in all stages of disaster management including preparation stages for pre-disaster periods, reaction to disaster during its occurrence as well as the post-disaster reconstruction is very important. Therefore, the capacity of IoT in creating intelligence in objects to investigate different environmental parameters, the capacity of analyzing information, the intelligence required to predict some events and consequently, responding timely to any event and even proposing programs and systems for human performance are very effective in managing disasters. The existing resources for investigating the usages of Internet of Things (IoT) in disaster management could each address a particular aspect of this application. This paper has attempted to present a comprehensive view regarding the different dimensions of application of IoT in disaster management. The advantages and disadvantages of this application are mentioned, some instances of previous systems and models have also been discussed. The novelty of this work is the extent of examined dimensions and also addressing what can be done in different kinds of disasters.

In this paper we present the study which uses hashing as a vectorizer and locality sensitive hashing to approximately find similar items, combined with incremental clustering to implement a practical real-time event detection algorithm. By gathering a substantial amount of Persian tweets, the proposed algorithm is evaluated. It is shown that the presented pipeline and methods are capable of detecting the events related to 7 out of 10 football matches during the days in which the Iranian national football team took part in the 2018 FIFA World Cup. A total of 102 events were detected with a precision of 87. 25%.

Several frameworks have been developed to evaluate the websites, especially their quality and credibility. The present study aimed to explore main categories of website evaluation such as source, content, infrastructure, and user. The main differences among the previous frameworks were identified by using a descriptive comparative method. The frameworks were divided into IS-oriented and user-oriented categories, which involved 23 and 6 models, respectively. Based on the findings, the technical aspects of the website were emphasized more in the field of computer sciences. In addition, the quality of the transmission process is discussed from source to audience in the communication discipline. Finally, regarding the field of information science, content and source criteria were considered more compared to other elements.

Steganography is the task of embedding a secret message in a cover media (e. g. image, video and voice) such that the stego media is not intuitively separable from the original ones. It is one of the interesting fields of security especially in transferring digital media as the major components of the web sites, emails and any web-based communications. Matrix embedding is a general approach used in many steganography schemes, especially where the cover size is small e. g. Voice Over IP (VOIP) packets. In this approach, the message is mapped to a series of bits (stego) to replace the same number of low significant bits in the cover. The embedding matrix is used for extracting the message by a linear combination of the stego bits. For a given matrix and secret message, there are specific series of stego-bits from which the message can be extracted. Embedding is done by an inverse problem to minimize a cost as the difference between the stego and the original bits. Hence, each message has a specific embedding cost based on the matrix. In this paper, a method is proposed to find an embedding matrix which minimizes the expectation of embedding cost for a uniform distribution of messages. To that end; a dynamic programming algorithm is proposed to efficiently find the expected cost for a given matrix. By use of this algorithm, any search method may be used to solve the problem. In this paper, a fast Hill-climbing search strategy is designed to find a local optimum matrix in an allowable time.

Everyone may influence others to change or persist in their current opinions via face-to-face or online communications. Predicting a society’ s majority opinion about a specific topic is an interesting challenge with many applications, e. g., predicting social movements, political votings, economical marketing. Among the various opinion formation models, the social impact model of opinion formation is very suitable for online social networks and online communities. In this model, three main factors affect a society’ s overall opinion: (1) the initial population of opinion groups, (2) the noise of the individuals to be persuaded or persist on their opinions, and (3) the topology of the network of interactions among individuals. In this research, to analyze the effect of segregation on the dynamics of opinion in the model, we assumed a noise-free model. Furthermore, the network of individuals is a scale-free network, and the initial population size of both opinion groups are the same with randomly assigned opinions. Using an agent-based modeling approach, we studied how the segregation of opinion groups may affect the dynamics of opinion formation. The results reveal that there is a strong correlation between segregation and the trend of society’ s opinion. It could be concluded from the results that if starting from the same population size in both opinion groups, it is expected that the more segregated opinion group dominates the less one and determines the majority opinion of the society.

Learners’ behavior analyzing in a learning environment and implicit discovery of learners’ personal characteristics has been as an interesting studies in recent years. In this method, since the learners’ personality characteristics takes place over time, we will obtain highly accurate personality information. In this paper, we will identify some aspects of Elliot’ s orientation goals using frequent pattern mining among information of changing/non-changing teammate of 92 students, extracted from a dynamic collaborative learning environment. The results of this study can be used in any adaptive system that needs to identify orientation goals based on its behavior to adapt learning system to learner’ s characteristic.

Task scheduling is the main challenge for the service provider in cloud Computing. One of the most critical objective in the scheduling is to assign tasks to virtual machines so that some machines do not overload or under load. To do this, load balancing plays a crucial role in the scheduling problem. Using an appropriate load balancing method can reduce response time and increase resource utilization. In this paper, we present a dynamic method for scheduling a task to virtual machines to increase load balancing and reliability in cloud computing. The proposed method reduces the makespan, increases the degree of load balancing, and improves the system`s reliability. The proposed method, in contrast to previous work, has been able to increase the reliability of the task scheduling based on previous experience of the virtual machines in addition to the fair distribution of workload among the virtual machines. We have compared the proposed algorithm with other task scheduling algorithms such as the honeybee load balancing and dynamic scheduling without load balancing. Simulation results show that the proposed method improves the reliability and degree of imbalance.

The day-to-day growth of data in organizations provides an opportunity for organizations that are capable of data analysis to gain competitive advantage. One of the industries where data analytics is very important is the banking industry; in this research, in order to evaluate the readiness of banks to use big data, a questionnaire whose content validity is reviewed by experts and reliability was confirmed by Cronbach's alpha test. The statistical population of the research is the IT experts in the banking sector and the sampling method is available sampling. The research method was descriptive-survey and one sample t-test was used to test the hypotheses. The results show that readiness at both data levels and data analysts and scientists is above average and lower than average at four levels of organizational, leadership, technology and goals.

Nowadays, guiding people toward correct and real information in a world full of internet obscurity is considered a highly important challenge and an urgent need. The present work aims to propose a validation method to those looking to discover the truth and verify the announced news using the knowledge and information of trustworthy people. In this method, the trustworthy people will first be identified based on the user-defined topic, then, it is discussed with them to determine if they have spoken about. Eventually, the similarity between the user-generated content (UGC) and trustworthy people-stated words in measured using the BERT method, and then, the system determines the extent to which UGC is false or true.

Nowadays, social networks have gained a lot of popularity among people. With the growth of these networks and a large number of people using these networks, social network analysis has received special attention, so the need for highly accurate and fast algorithms on various issues is strongly felt. One of the important issues in these networks is community detection problem that many algorithms have been proposed for this purpose. In social networks, communities usually are formed around popular or influential nodes. Most algorithms in this field, that are usually density-based, are unable to detect this structure. In this paper, we propose a new community detection algorithm based on the local popularity structure. In this algorithm, the most popular person in neighborhood of each user is selected as a leader and the user falls into that group. Experimental results on six real networks show that the proposed method not only has comparable results in terms of NMI and ARI, but also has shorter execution time compared to existing algorithms.

Connecting your system or device to an insecure network can create the possibility of infecting by the unwanted files. Malware is every malicious code that has the potential to harm any computer or network. So, detecting harmful files is a crucial duty and an important role in any system. Machine learning approaches use a variety of features such as Opcodes, Bytecodes, and System-calls to achieve accurate malware identification. Each of these feature sets provides a unique semantic view, while, considering the effect of altogether is more reliable to detect attacks. Malware can disguise itself in some views, but hiding in all views will be much more difficult. Multi-View Learning (MVL) is an outstanding approach that considers multiple views of a problem to improve the overall performance. In this paper, inspiring MVL an approach is proposed to incorporate some various feature sets and exploit complementary information to identify a file. In this way, the consensus of multiple views is used to minimize the overall error of a classifier based on sparse representation. To show the generalization power of the proposed method, various datasets are employed. Experimental results indicate that in addition to high performance, the proposed method has the advantage of overcoming the imbalanced conditions.

AdaBoost is perhaps one of the most well-known ensemble learning algorithms. In simple terms, the idea in AdaBoost is to train a number of weak learners in an increamental fashion where each new learner tries to focus more on those samples that were misclassfied by the preceding classifiers. Consequently, in the presence of noisy data samples, the new leraners will somehow memorize the data, which in turn will lead to an overfitted model. The main objective of this paper is to provide a generalized version of the AdaBoost algorithm that avoids overfitting, and performs better when the data samples are corrupted with noise. To this end, we make use of another ensemble learning algorithm called ValidBoost [15], and introduce a mechanism to dynamically determine the thresholds for both the error rate of each classifier and the error rate in each iteration. These threshholds enable us to control the error rate of the algorithm. Experimental simulations have been made on several benchmark datasets including Web datasets such as “ Website Phishing Data Set” and “ Page Blocks Classification Data Set” to evaluate the performance of our proposed algorithm.

Recent advents of the internet have made accessibility of people to digital data such as audio, images, and videos much easier. Meanwhile, one of the cases that adversaries take advantage of is the people's face images that are available across the web. Digital watermarking is used to authenticate the original owner of the images and protect their copyright. With the help of digital watermarking, hidden data is embedded inside the image. Recently, neural networks such as autoencoders are one of the most popular models that are used in many fields. Neural networks are capable of understanding all kinds of raw data such as images and videos. In this paper, we present a method for embedding the user's national ID in their face images using autoencoders. The proposed autoencoder is trained with a dataset contains face images. The image is coded into some code using the autoencoders' encoder. Then, the national ID is embedded in this code and the modified code is reconstructed using the decoder to form the watermarked image. To extract the watermark, the watermarked image is encoded with the encoder and the watermark is extracted. Experiment results show that our model recovers the watermark with high accuracy and it is resistant against JPEG attacks. Moreover, the quality of the watermarked images is acceptable, and their SSIM compare to the original image is about 90%.

Tourism is one of the most important industries in the world. It plays an important role in the cultural and economic prosperity of countries. Therefore, the design and production of tourism recommendation systems has become one of the most popular topics for researchers in order to satisfy tourists. In this paper, a tourism recommendation system is proposed. In this system, tourists' interests are extracted by semantics using word clustering and emotion analysis. Finally, recommendations based on the interests of tourists are presented to them. The evaluation results of the proposed system provide high values in the evaluation criteria.

With the increasing rate of online information generation and the growth of Internet users, the need for automatic text summarization has become more urgent. In this paper, a multi-document summarization method based on learnable submodular functions is provided. Using submodularity, it is possible to guarantee the quality of solution. Input text documents are modeled using weighted graphs, where nodes and edges represent sentences and similarities between them, respectively. Using this graph, the features that indicate the significance and impact of each sentence are extracted. Significance features consider the value of any sentence, independent of other ones. Modular functions are used to model these features. On the other hand, impact features consider the relationships between sentences. For modeling these features, submodular functions are used. These features are the building blocks of the target function which is equivalent to an ordinary neural network. Therefore, using a training set, network learns how to summarize text. After learning phase, the obtained function is used to summarize input texts. The summarizer has been examined using Pasokh and DUC 2004 datasets, and its results are presented.

Today millions of web users put their opinions on the internet about various topics. Development of methods that automatically categorize these opinions to positive, negative or neutral is important. Opinion mining or sentiment analysis is known as mining of behavior, opinions and sentiments of the text, chat, etc. using natural language processing and information retrieval methods. The paper is aimed to study the effect of combining machine learning methods in a meta-classifier for sentiment analysis. The machine learning methods use the output of lexicon-based techniques. In this way, the score of SentiWordNet dictionary, Liu’ s sentiment list, SentiStrength and sentimental words ratios are computed and used as the input of machine learning techniques. Adjectives, adverbs and verbs of an opinion are used for opinion modeling and score of these words are extracted from lexicons. Experimental results show that the meta-classifier improve the accuracy of classification 0. 9% and 1. 09% for Amazon and IMDB reviews in comparison with the four machine learning techniques evaluated here.

The main purpose of data mining is to discover hidden and valuable knowledge from data. The Apriori algorithm is inefficient due to bulky deals of searching in a dataset. Bearing this in mind, this paper proposes an improved algorithm from Apriori using an intelligent method. Proposing an intelligent method in this study is to fulfill two purposes: First, we demonstrated that to create itemsets, instead of adding one item at each step, several items could be added. With this operation, the number of k-itemset steps will decline. Secondly, we have proved that by storing the transaction number of each itemset, there would be a diminishment in the time required for the dataset searches to find the frequent k-itemset in each step. To evaluate the performance, the Intelligent Apriori (IAP) algorithm has been compared with the MDC algorithm. The results of this experiment exhibit that since the transaction scans used to obtain the itemset momentously reduced in number, there was a considerable fall in the runtime needed to obtain a frequent itemset by the proposed algorithm. In this study, the time required to generate frequent items had a 46% reduction compared to that of the MDC_Apriori algorithm.