کنفرانس بین المللی وب پژوهی
سال:1403 | دوره:10 |تعداد مقالات:80

This paper presents an efficient method with simplified calculations for classifying left and right hand movement imagery using a single channel of EEG to design a brain-computer interface (BCI) system. The proposed method utilizes wavelet transform to decompose EEG signals into multiple frequency bands. The classification task is performed using the support vector machine (SVM) with the radial basis function (RBF) kernel and the K-nearest neighbor (KNN) algorithm. This paper introduces and compares two techniques that are based on the sixth and eighth levels produced from a deconstructed single C3 channel. For feature extraction, statistical information such as variance, standard deviation, and signal power are individually considered. The achieved results indicate high accuracy for left and right-hand movement, with 100% accuracy for left-hand movement and 87. 47% accuracy for right-hand movement. These results were accomplished using SVM with the RBF kernel and KNN algorithms, based on power features extracted from the eighth-level EEG signal. Compared to prior methods utilizing single-channel and multi-channel approaches, this method demonstrates superior performance.

According to the research conducted, people spend about 70-90% of their living and working time indoors. Therefore, providing systems that offer adequate services to users in these environments seems essential. Locating users and devices is widely used in healthcare, industry, building management, surveillance, and other areas. There are various technologies for indoor positioning systems. In this paper, Ultra Wide Band (UWB) technology is considered due to its high accuracy in indoor positioning. However, there are many objects and people in indoor environments, so obstacles can reflect the transmitted signals. Compared to the Line of Sight (LoS) signal, the delay of the signal transmission path in the Non-Line of Sight (NLoS) signal leads to positive range errors. In order to reduce the effect of NLoS conditions on positioning. In this research, we have attempted to achieve high-precision accuracy separation for LoS and NLoS conditions by providing deep learning networks and using channel impulse response data as input without prior knowledge of the environment. In addition, the result of this classification is compared to other references that used a similar dataset. The results of the NLoS/LoS signal classification section show that the proposed Convolutional Neural Networks (CNN) are better than other neural network methods (such as Deep neural networks).

Graph neural networks have gained a great popularity in the past few years because they have proven to be useful in many tasks in complex networks, including link prediction. The complex and multi-layered structure of multiplex networks poses challenges to traditional link prediction methods. In this study, we propose a new approach based on Graph Neural Networks (GNN) for link prediction in multiplex networks. In the suggested approach, several adjacency matrices have been aggregated based on measuring the inter-layer similarities and employed in a GNN. The experimental results on benchmark real-world networks show the effectiveness and validity of the method.

In the era of next-generation wireless networks, real-time applications demand freshness in information delivery, notably in IoT, cyber-physical systems, and industrial IoT. We target the optimization of average Age of Information (AoI) in a multi-user RIS-assisted mmWave network, where the sensitivity of mmWave bands to propagation challenges is mitigated through reconfigurable intelligent surfaces (RIS). Notably, prior studies have overlooked AoI optimization in RIS-enhanced mmWave networks. Our contribution lies in devising a novel Markov Decision Process (MDP) framework and a model-free Deep Reinforcement Learning (DRL) algorithm tailored to high-dimensional action spaces. Our approach orchestrates user transmission timing, power allocation, base station beamforming, and RIS reflective coefficient adjustments. Through extensive simulations, we demonstrate the superiority of our method over existing schemes, showcasing enhanced network performance. This work represents an advancement in the optimization of AoI within the context of RIS-assisted mmWave communications, shedding light on efficient control policy learning for improved network operation.

The aim of increased reliability and performance in system architectures has resulted in the development of data replication systems. However, preserving consistency during concurrent requests across various replicated data servers presents a significant challenge, especially when write operations are involved. This issue highlights the necessity of establishing a precise balance between system performance and consistency, particularly in the context of large-scale distributed databases. Large-scale distributed databases must navigate a trade-off between performance and consistency. Client-centric (CC) Consistency addresses this problem by guaranteeing consistency for single-client access to distributed data stores (DDS). CC consistency encompasses four distinct types of consistency assurances: 1) monotonic read, 2) monotonic write, 3) read your writes, and 4) write follow reads. In this paper, we present a formal model encapsulating CC consistency and its associated consistency models within distributed systems. Our model uses high-level ML functions to represent CC consistency and its models. Integrity is verified CPN tools, ensuring consistency. We demonstrate its effectiveness in discerning supported CC consistencies with relevant inputs. This formal approach aids understanding and assists in optimizing distributed system performance while maintaining consistency.

The utilization of computer systems has rapidly expanded, accompanied by a corresponding rise in security threats such as hackers, viruses, worms, and similar malicious entities spreading at an alarming rate across networks. In response, anomaly intrusion detection methods have been developed to counter these attacks. However, as information systems evolve, certain detection techniques have seen a decline in effectiveness due to the escalating volume of network data traffic and the continuous need for swift responses. Addressing this critical issue, this research proposes a method to enhance the accuracy of feature selection and extraction for intrusion detection and anomaly classification. This is achieved through the integration of optimization and autoencoder algorithms, evaluating the impact of machine learning and artificial intelligence in network anomaly detection. Utilizing the NSL-KDD dataset, the study begins with data collection and preparation, followed by the application of optimization algorithms such as the Rain Optimization Algorithm (ROA) and Artificial Bee Colony (ABC) in conjunction with various neural network architectures, including Radial Basis Function neural network, decision tree, Support Vector Machine, K-Nearest Neighbors, ensemble network, mountain model, SOM clustering, and ultimately the Hoeffding Tree-based Autoencoder network. Results demonstrate that the proposed approach, leveraging the Rain Optimization Algorithm and Hoeffding Tree-based Autoencoder network, excels in feature selection and extraction during training, effectively detecting and classifying intrusion or anomaly occurrences with high accuracy. Notably, among the algorithms tested, the Hoeffding Tree-based Autoencoder network achieved an accuracy of 98. 74%, indicating commendable performance in classification and result evaluation.

The application of blockchain technology in the context of the General Data Protection Regulation (GDPR) has emerged as a significant and challenging area of research in recent years. Many believe that blockchain capabilities can be leveraged to comply with GDPR principles. This study focuses on the specific capabilities of blockchain technology and addresses the unique data protection challenges in the information age, aiming to identify and represent these relationships through an applicability model. The goal of this research is to present a model for evaluating and selecting blockchain applications that are aligned with GDPR requirements. This model is developed through a meta-synthesis method and qualitative content analysis of 67 articles that concentrate on the potential applications of blockchain in GDPR across various countries. These applications are defined based on an analysis of key blockchain capabilities such as transparency, immutability, and data encryption. Further, this study maps these capabilities to key GDPR requirements such as data minimization and accuracy, analyzing them from both facilitating and inhibiting perspectives. Ultimately, based on qualitative content analysis and an in-depth study of 31 articles selected through the Critical Assessment of Methods of Protein Structure Prediction method (CASP), this research offers a comprehensive model for understanding and leveraging blockchain technology in a manner that complies with GDPR requirements. according to this, the study contributes to the understanding of blockchain technology’, s applicability in meeting GDPR requirements, offering insights for policymakers, industry stakeholders, and technology developers on the strategic implementation of blockchain for data protection.

There are studies on whether there may be a relationship between academic citation and web visibility. The purpose of this study was to more accurate about this relationship. Search Engine Optimization is the technique for receiving high visibility in the web environment. Academic SEO, due to the impact of open science, is extending. The uniform resource locators of the 80 open-access journals indexed on Elsevier that had two benchmarks, Cite-Score, and Impact-Factor were analyzed. For extraction of SEO-Scores of websites of these mentioned journals, used Seobility automatic analytical tool (October 2022). The data analysis tools were SPSS version 22 and Excel software. There was no significant correlation between the Google SEO ranks of websites of the open access journals and their Cite-Scores. In addition, between their Impact Factors, and SEO scores, no relationship was seen by the Spearman correlation test. Therefore, most of the citations are based on deep knowledge rather than web visibility, most probably. In other words, this current article indicates that the deep web, or what scientists do not see on the first pages of their results when they search and retrieve and cite, but on the next pages, seen and then have cited to them, is still important. This result does not contradict studies that show a correlation between citation and visibility. But, only suggests the existence of many factors in the citation of academic articles other than their open-access state, such as the deep web and citing academic celebrities.

Today, autoscaling in cloud environments is critical and fundamental. This concept refers to a system's ability to adapt to traffic fluctuations and variable user needs without human intervention. Given the rapid changes in demand in the field of information technology, autoscaling serves as a reliable and essential tool, assisting organizations in facing informational market challenges and enhancing system flexibility and efficiency. While existing methods and algorithms have made significant advances, there are deficiencies that require investigation. The complexity and time-consuming nature of algorithms, the need for precise configurations, instability when facing sudden workload spikes, and low accuracy in predicting future workloads are among the issues that necessitate study and research. In this study, an approach for cluster autoscaling in cloud environments based on workload prediction is presented, utilizing a multi-stage hybrid model. This method predicts workload using initial prediction models and enhances the final prediction accuracy and consequently scalability performance through weighting and combining these models. The proposed algorithm in this research has been experimented and evaluated using Google cluster data, showing that it has achieved a precision of 0. 99 and an error rate of 0. 05 compared to the baseline accuracy of 0. 56, indicating an increase in prediction accuracy by 0. 34 and better performance with higher accuracy.

Considering the need for knowledge sharing in various semantic web domains, the concept of ontology has been introduced as a method for standardizing information. However, for collaboration between different knowledge domains, the need for alignment between ontologies has arisen. Ontology alignment refers to the process of mapping pairs of entities between two ontologies. The most important component for finding aligned pairs of entities between two ontologies is the use of similarity measures. Similarity measures encompass three categories: semantic, lexical, and structural similarities. Some alignment methods use a single measure, while more advanced approaches like AML and LOGMAP combine multiple similarity measures to achieve better results. In this article, we have been able to improve the AML method by emphasizing the importance of the grammatical position of words in the textual content of entities, and enhance its performance to 94% in terms of the F-measure in the anatomy ontology dataset.

With the rising prominence of gold as a lucrative investment avenue in Iran, this research delves into predicting the future price of 18-carat gold. In pursuit of this objective, a comprehensive comparison is conducted between two neural network architectures: the Gated Recurrent Unit (GRU) as a single structure and a hybrid model combining Convolutional Neural Network (CNN) and Long Short-Term Memory Neural Network (LSTM). The evaluation criteria employed focus on error metrics to gauge the accuracy of price predictions. Results reveal that the CNN-LSTM hybrid neural network exhibits superior performance, showcasing lower error values in predicting the price of 18-carat gold in Iran. Consequently, the chosen model, CNN-LSTM, is employed to forecast the following day's gold prices, providing valuable insights for investors navigating the Iranian market. This research contributes to the ongoing discourse on gold investment strategies by highlighting the effectiveness of advanced neural network models in enhancing predictive accuracy.

The wide range of models employed in many categories is constantly growing due to advancements in Model-Based System Engineering and Model-Driven Engineering, as well as their acceptance in academic and industry communities. Consequently, intelligent approaches have to be developed to manage these models, including comparing, clone detection, analyzing, and searching for similarity. There are obstacles and restrictions when using machine learning algorithms on models because of their special structure. While these models comprise graph structures, they also contain textual information. Both of these features must be considered to perform more accurate learning. This paper explores the challenges associated with this area of study and discusses potential solutions. Since comparison is the basis of many learning algorithms, it is imperative that the data in this category be prepared in a fashion suitable for comparison. In this sense, the text and structural aspects of embedding meta-models (as one of the popular techniques in the model’, s world) techniques are assessed, and an approach for combining both aspects is suggested. Studies in this area have been carried out using evaluation criteria including goals, base method, efficiency, accuracy, and scalability. The outcomes indicate a text-structure combination can work well as an embedding model for more investigations.

A significant amount of interest has been generated in recent years in the convergence of quantum computing and data mining due to quantum algorithms' potential to revolutionize information extraction from vast datasets, as well as their ability to utilize quantum principles and natural capabilities to perform far more efficiently and efficiently than classic algorithms. We provide a comprehensive and detailed review of the current state of quantum computing algorithms in the field of data mining and machine learning, highlighting key developments, challenges, and their implementation. In particular, machine learning can benefit from QC mechanics like superposition and entanglement due to the use of q-bits as well as other superior advantages provided by quantum computing, such as Grover algorithm. It is important to note that classic machine learning models differ in terms of the types of learning, prediction models, and operating on data as well as performance over time, along with their counterparts, such as supervised and unsupervised models, which are naturally more sophisticated. Our review will also discuss how integrating and merging the two fields of machine learning and quantum computing, and their mechanics, will affect the aspect of time and resource.

The present research deals with the approach of expanding the interactive aspect of digital advertising in the environment and the structure of its communication process with the audience, with the aim of investigating the methods of interactive advertising in relation to the way it communicates with the audience. Today, this method of advertising is due to the way the audience participates in completing the advertising message and the attractions it has for the audience,It is increasingly growing and expanding. Many companies, educational institutions, health and social institutions try to attract and participate the audience to introduce their products or message. Accordingly, the purpose of such a research is how to benefit from audience participation, the aspects of interaction in digital interactive advertising in the environment and the methods of sending and receiving messages. The purpose of this research is the process of creating a relationship between the audience and interactive advertising. This research has been done in a descriptive-analytical way and based on the approach of digital advertising in the environment and the way of audience participation based on audience theory. In conclusion, the results of the research show that interactive advertising should be done by creating a sense of curiosity and innovation for the audience, creating visual appeal, presenting the message based on the personality needs of the audience, so that the audience is encouraged towards the advertising message. In this regard, by examining foreign and Iranian examples of interactive advertising, we have analyzed and explored the components, requirements, features and method of persuading the audience.

With the widespread use of social media, communication, and news dissemination have become faster and more extensive. Therefore, identifying key individuals in controlling rumors is of great importance. Researchers have employed various methods to identify these individuals, each with its advantages and disadvantages. Some methods lack accuracy and do not adequately identify the influence of each individual. Others have low resolution, meaning they consider many individuals to be equally influential and do not differentiate between them. In this paper, we propose a comprehensive method based on the Degree and K-shell, which improves upon the Hybrid global structure model (HGSM) method. According to the evaluations performed on five social networks, the proposed method has higher accuracy and resolution compared to recent methods, while also having a suitable time complexity. The real influence of each individual is calculated using the Susceptible-Infected-Recovered (SIR) model and the evaluation criteria are Kendall's Tau, Monotonicity index, and Time complexity.

The purpose of this study is to examine the mediator role of Accounting Information Systems (AIS) between organizational Culture (OC) and Organizational Performance (OP) in Small and Medium Enterprises (SMEs). This study surveyed SMEs employees in Iraq, which were selected using the convenience sample technique. The data were tested for validity and reliability, and finally, 212 valid questionnaires were analyzed using Smart-PLS. The results indicated that the AIS mediated the relationship between OC and OP, while OC and AIS significantly impacted OP. This study also found a significant impact of OC on AIS. The SME managers and owners should pay more attention to the importance of AIS and develop the culture that will lead to implementing the AIS to enhance OP. Overall, the study enriches the body of knowledge and extant literature by examining the OC, AIS and OP among SMEs in developing countries, particularly Iraq.

Purpose: This article is designed to provide visual solutions to increase the efficiency of online advertising banners, by examining the relationship between visual factors and user attraction through clicking on ads. Method: From the point of view of the application goal and from the point of view of the research method, the current research method is a survey-analytical description, the current state of Internet advertising banners and providing appropriate visual solutions using previous sources and available data have been investigated. Findings: The research findings showed that there is a positive and significant relationship between visual factors and targeted clicks, and a good design strengthens the message and attracts attention, and this attention will increase the click rate. Conclusion: The results of the findings showed that motion and CTA key, two important factors in internet advertising, have a direct and meaningful relationship with the user's behavior and his interest in clicking, as well as the element of color, on the relationship between the audience and persuasion to click and redirect to the page. Landing has a moderating effect.

Function-as-a-Service (FaaS) allows to directly submit function code to a cloud provider without the burden of managing infrastructure resources. Each cloud provider establishes execution time limits to their FaaS offerings, which impose the risk of spending computation time without achieving partial results. In this work, a framework that enables limitless execution time in FaaS, with little to no modifications to the user-provided function code, is presented. After a thorough literature and theoretical framework review, Apache OpenWhisk Actions and the DMCTP checkpoint-and-restore (CR) tool were selected. With these, dependent successive serverless same-function invocations that exploit the persistence of partial results were implemented. The solution was submitted to the FaaSDom benchmark and time metrics were collected. Additionally, the solution was characterized in terms of the Serverless Trilemma. The resultant system, even at this proof-of-concept state, offers a lot of value to companies that rely heavily on serverless architecture.

Graph representation learning aims to extract embedding vectors for graph nodes, such that similar nodes have close vectors in the embedding space. Existing methods often measure node similarity based on their common neighbors, which may overlook nodes with similar structures in different parts of the graph. We want to capture the structural similarity of nodes that are not adjacent in the graph. To this end, we propose struc2vec+k, a new method that extends the basic struc2vec method. The basic method considers two nodes to be structurally similar if their nodes in the first, second, third, and subsequent layers are similar. The proposed method also takes into account the connection between layers, and aggregates the information of two consecutive layers. For instance, for the second layer, the information of the first-and second-layer nodes are aggregated. This aggregation is based on the inter-layer connections. The aggregation can be done up to the k-th layer, which explains the name of the method. We show that the proposed method achieves good accuracy in numerical experiments.

Some networks are well-modelled as a multilayer structure in the real world, with interactions between nodes in numerous layers. For example, users may, have accounts on numerous online social networks (OSNs) such as Twitter, Instagram, and Facebook, and each social network can be thought of as a layer in a multiplex network. The goal of predicting links between layers of a multiplex network is to detect if accounts across various OSNs are associated with a same user or not. This can be useful in a variety of situations, such as evaluating client interests or predicting cybercriminal behavior. In this research, we present a unique Interlayer link prediction approach that incorporates information from the degree penalty mechanism. The technique takes advantage of the network's power-law degree distribution. As a result, in different OSNs, neighbors with varying degrees of proximity may have varied effects on the degree of node matching. It can match a user's accounts across many layers well. The suggested strategy outperforms similar methods by at least 10% in terms of prediction accuracy, according to experimental results on both synthetic and real-world networks.

In this article, the problem of learning representation in heterogeneous graphs is investigated. Due to the presence of different types of nodes and edges in this type of graphs, there are unique challenges that limit the possibility of using conventional graph representation techniques. The way of random walk in this type of graphs is different and they need a walking scheme or metapath to find the path. Specifying this scheme is one of the challenges of learning representation in heterogeneous graphs. In this article, an algorithm has been introduced that finds all possible metapath schema by taking an heterogeneous graph and finds the best metapath scheme by specifying the correct schema and checking them. Various experiments show that with a small sampling of the network in the form of short length, the most suitable scheme can be found automatically and it is shown that by changing the sampling size, the selected scheme is the best scheme and in terms of time only Runs in 0. 007% of the time using long random walks.

The increase of traffic in Internet-of multimedia Things networks leads to additional load on servers,therefore, this paper focuses on server load balancing in multimedia Internet-of-Things networks. Software-defined networking technology has been used to achieve load balancing in these networks, as software-defined networks with new features have improved load balancing in multimedia Internet-of-Things networks. In this study, the short-term and long-term recurrent neural network algorithm is used to predict the server load, and then a fuzzy system is used to accurately determine the server levels. Also, this article saves energy and also reduces server overhead.

This research explores indoor location estimation in the Internet of Things (IoT) using Wi-Fi signals. Due to the influence of surroundings on signal strength, advanced techniques like Machine Learning (ML) algorithms are needed for accurate location estimation. The challenge lies in balancing accuracy with real-time operation and resource limitations. This research proposes using Ensemble Learning (EL), combining multiple machine learning models, to improve accuracy. However, this increases computational cost. The study uses Genetic Algorithm (GA) to find an optimal Ensemble model that balances accuracy and resource usage. The proposed method is tested on Wi-Fi data collected across multiple floors of a building, unlike previous studies that focused on a single floor. This comprehensive approach leads to an average localization error of 0. 27 meters with an accuracy of 95%. In comparison to previous single-floor studies, the model achieves an accuracy of 98. 3% with an error of 0. 25 meters on it.

Stance detection, which refers to the assessment of a statement's position regarding a specific target, is recognized as an important area in natural language processing. This process generally involves identifying whether a claim is in favor of, against, or neutral towards a particular subject. With the expansion of social networks and the increase in the sharing of opinions and viewpoints, stance detection from textual data has become a powerful tool for analyzing and understanding public opinion. To date, most research in this field has focused on using trained word embedding models to assess the stance of texts. However, these efforts often overlook the key role of the targets against which a text is evaluated. In this paper, considering the importance of determining targets in stance detection, we present an innovative approach that focuses on target-based embeddings. Using the SemEval2016 dataset, which includes five different targets, allowed us to demonstrate the effectiveness of our proposed method. We were able to achieve an average accuracy of 79. 5 percent on this dataset, which is a 4 percent improvement over the results of previous works.

In the contemporary period, the acceptance of medicinal plants and herbal medicines is on the rise due to factors like accessibility, reasonable cost, and the belief in lower side effects compared to chemical drugs. Despite common belief, medicinal plants have side effects, and their misuse can result in irreparable damage. Given these challenges and the lack of appropriate knowledge structures, we initiated the development of the first Herbal Medicine Ontology of Iran (HMOIr). This ontology manually curated using data from GolDarou, a leading herbal pharmaceutical company in Iran, comprises 971 classes, 16 relations, and 4 annotations. Evaluation, comparing with primary sources and competency questions, revealed coverage and accuracy percentages of 89. 17% and 97. 2%, respectively. Lastly, the Iranian Herbal Medicines Information System, built upon this ontology, was introduced to provide users with accurate responses to their inquiries.

Today, electronic medical records (EMRs) are generally kept in centralized databases. Hospitals generally have control over EMRs. It is a problem that EMRs are not kept decentralized and patients cannot manage their data. It is important that EMRs are kept confidential and secure in databases that can be accessed remotely. In addition, control of a patient’, s EMR should rest with the patient themselves. In this study, a model for sharing, accessing and storing EMRs without security vulnerabilities was proposed. The three operations (data saving, data sharing and hash control) for the proposed model are implemented. Encryption and hash algorithms with strong keys are used for secure data communication in the operations. A mobile application was developed according to this model. In the developed application, the entire authority over a patient EMR lies with the patient. EMRs are stored encrypted in the document-based database. The EMR digests are stored in the blockchain. In this way, a blockchain and a distributed database are successfully used together in the proposed model. The potential impact of the study is remarkable, as it provides data security without the need for a central authority.

Nowadays, in many developing countries, the surge in urbanization and vehicle congestion has led to a rise in traffic density and city accidents. This research aims to explore and compare various advanced traffic detection methods, accident prediction in smart cities using image processing, deep learning algorithms, and Internet of Things (IoT) technologies. These technologies include infrared sensors and video image processors installed on roads, and accident prediction systems. The goal is to alleviate traffic congestion, enhance vehicle flow, and decrease accidents. The study concludes with a comparison of the reviewed algorithms’,accuracy and speed in traffic detection and accident prediction.

The increasing growth of social networks has drawn researchers' attention to link prediction, and it has been used in many fields, including computer science, information science, and anthropology. One of the newest link prediction methods is graph embedding methods, which are used to generate a feature vector for each node of the graph and find unknown links. The DeepWalk algorithm is one of the most popular graph embedding methods that captures the network structure using a random walk with equal probability. In this paper, a modified version of the DeepWalk algorithm is proposed, which uses a new random walk model to solve the link prediction problem. In fact, in the proposed method, the amount of structural similarity and the similarity of important features of nodes are combined. The results show that two nodes are more likely to form a link if they have similar structure and important features. To evaluate the proposed method, experiments have been conducted on five datasets. The test results indicate a relative improvement in the results obtained.

Medical professionals use low-dose computed tomography (LDCT) to reduce radiation exposure in patients, but this can create noisy images and artifacts that complicate interpretation. Recent research focuses on using deep learning techniques to improve image quality in LDCT scans. In this study, we suggest a new method that combines EfficientNetV2 with a generative adversarial network (GAN) using Wasserstein distance and perceptual similarity. This approach helps reduce noise while maintaining LDCT image structures, potentially enhancing diagnostic accuracy and patient safety. By integrating EfficientNetV2 with a GAN and utilizing perceptual similarity and Wasserstein distance, we achieved excellent results with a PSNR of 32. 6058 and SSIM of 0. 9135 on the AAPM-Mayo Clinic LDCT Grand Challenge Dataset. The significant improvement over existing methods highlights the potential of our proposed method in enhancing LDCT image quality.

Images are among the most used components used in websites, and if these images are not of good quality, the website will suffer from SEO. Rainy images are among the poor-quality images that inevitably exist in the display of products on a store website. Rain removal methods using weighted average filters have the disadvantage of blurring images and destroying image details. The innovation of this article is to propose a combined method of Bilateral filter and Guided filter to remove the effect of rain. The working method is that first the image passes through the Bilateral filter, and in order to increase the quality, the guided filter is applied to it and the output image is obtained. Four sample images are selected from rain images and the proposed algorithm is applied to them. The results have been compared with a reference method recently presented in this regard. The results show that the proposed algorithm has improved the output of the images by 4. 42% on average in terms of the maximum signal-to-noise criterion and 4. 45% in terms of the similarity criterion compared to the reference algorithm.

Recently, graph data has received a lot of attention as it is used to represent other data types including social networks, banking, security, financial, medical, and textual data. Therefore, anomaly detection in such data has become an important research area due to its ability to prevent adverse events like financial fraud, network intrusion, and social spam. Anomalies in graph data can occur at the node, edge, subgraph or graph level. Node anomalies can be due to unusual structure or attributes. Edge anomalies aim to identify unusual connections which are often unexpected or unusual relationships between real-world entities. The main challenge is detecting and categorizing such anomalies. In recent years, many computational methods have been proposed for anomaly detection in graphs using statistical analysis and machine learning approaches. In this research, we propose a framework called "GRAPH-Guard" for node-level anomaly detection using deep and ensemble learning algorithms. The framework is evaluated on heterogeneous attributed graphs and compared to other node anomaly detection algorithms quantitatively and qualitatively. The quantitative evaluation shows the proposed framework improves the area under the curve (AUC) by 4 points compared to the average AUC of previous works. It also improves the F1 score by 1 point compared to the best previous F1 score.

Today, social networks have attracted the attention of billions of Internet users. On the other hand, the widespread use of these networks is susceptible to many dangerous purposes, such as spreading malware, stealing user information, spreading false information, etc. In this article, the effective detection of bots in the Twitter social network is proposed using the deep Boltzmann machine, which is one of the important types of deep neural networks. Indeed, various methods have been provided to detect bots in social networks. It is essential to extract key features that directly impact the accuracy of the methods. In order to achieve this goal, the Boltzmann machine neural network has been developed to extract the key and important features from the bunch of features included in the Twitter dataset. Then, based on the selected features, bots are detected using different classification approaches such as the K-nearest neighbor, support vector machine, AdaBoost, and decision tree, which provide better performance than the existing methods.

Through social networks, which are groups of individuals and their relationships, people are often influenced by one another. Each individual in the network may propagate their behavior or ideas to those they are connected with. Thus, influence propagation occurs when a group of individuals exhibits a particular behavior or idea, and it spreads through the network due to interpersonal connections. Advertising, marketing, and public health can benefit from studying this phenomenon. The aim of this study is to pinpoint the most influential individuals in a social network so they can maximize their impact. As a result of the proposed method (DQVNS), the variable neighborhood search algorithm is improved by combining deep reinforcement learning (RL) and variable neighborhood search algorithms. Extensive evaluations on real social networks of diverse sizes confirm this algorithm's significant advantage over traditional heuristic approaches.

Open data is a new trend in which researchers and organizations have made their data freely and publicly available in recent years. Although open data offers several advantages, including transparency, service improvement, value chain creation, and development of businesses, it faces single point of failure challenge. Moreover, there is no solution to maintain integrity and trustworthiness of the open data presented. Blockchain technology is a secure platform for storing and retrieving open data to address these challenges, however, there are challenges to blockchain-based open data research that require further investigation. In this paper, firstly, we study several applications presented in the area of blockchain-based open data research, and then prepare a taxonomy of complementary techniques that have been applied to address the open data challenges. The proposed taxonomy prepares proper knowledge for organizations to decide which complementary techniques of open data storage and retrieval can meet their requirements. In conclusion, we show smart contracts and Internet of Things (IoT) are outstanding complementary techniques for data storage and retrieval that have been applied in a wide range of blockchain-based open data research.

This paper explores the capability of various machine learning algorithms, including Random Forest and advanced gradient boosting techniques such as XGBoost, LightGBM, and CatBoost, to predict customer churn in the telecommunications sector. For this analysis, a dataset available to the public was employed. The performance of these algorithms was assessed using recognized metrics, including Accuracy, Precision, Recall, F1-score, and the Receiver Operating Characteristic Area Under Curve (ROC AUC). These metrics were evaluated at different phases: subsequent to data preprocessing and feature selection,following the application of SMOTE and ADASYN sampling methods,and after conducting hyperparameter tuning on the data that had been adjusted by SMOTE and ADASYN. The outcomes underscore the notable efficiency of upsampling techniques such as SMOTE and ADASYN in addressing the imbalance inherent in customer churn prediction. Notably, the application of random grid search for hyperparameter optimization did not significantly alter the results, which remained comparatively unchanged. The algorithms' performance post-ADASYN application marginally surpassed that observed after SMOTE application. Remarkably, LightGBM achieved an exceptional F1-score of 89% and an ROC AUC of 95% subsequent to the ADASYN sampling technique. This underlines the effectiveness of advanced boosting algorithms and upsampling methods like SMOTE and ADASYN in navigating the complexities of imbalanced datasets and intricate feature interdependencies.

One of the most important aspects of learning is attention. This is even more pronounced in online learning due to the instructor’, s need for sufficient control over the learner’, s environment. This study’, s purpose is to identify the pattern of changes in the cognitive state of learners from unconsciousness to consciousness. By observing the brain’, s response while learners watch micro videos, we sought to understand the impact of these ego states on learners’,performance in E-Learning environments. Our findings suggest that learners' ego states significantly impact their learning performance. In the first phase, the purpose was to detect the transition point from the unconscious to the conscious state precisely. In the second phase, we tried to differentiate between these two states by comparing their learning performance. Finally, the obtained results led us to believe that learning outcomes are subject to a significant increase when the brain state changes.

Consumer acculturation pertains to the overall adjustment and adaptation to the cultural setting of a particular society by individuals originating from a different culture. On the other hand, when analyzing consumer purchases, consumer personality traits become an important aspect. Also, social media connects people from different parts of the world with different cultures and enables them to interact and communicate with each other. Therefore, this research was conducted to investigate the mediating role of social media in the influence of consumer personality on consumer acculturation. The statistical population of this research was 188 consumers of luxury brands who were members of online brand communities before purchasing and had relationships with other consumers. The findings indicated that extroversion, agreeableness, and openness to new experiences positively and significantly impact customers' acculturation process. The personality traits of neuroticism and conscientiousness did not have a positive and meaningful effect on the customer's acculturation. On the other hand, the use of social media plays a mediating role between all 5 personality traits and acculturation.

Today, data is more valuable to us than gold. When observing the environment, a substantial amount of data, particularly textual information, can be identified, tagged, prepared, and published in the form of a corpus or datasets. The primary objective of our paper is to gather, prepare, tag, and develop a vast dataset of Fidibo users' opinions regarding educational content and e-books. This dataset enables in-depth analysis of emotions and opinion mining, particularly within the educational content realm. A common flaw in nearly all similar datasets in the Farsi language is their restriction to user opinions on services and products available on online platforms. The dataset we refer to as LDPSA (A Large Dataset of Persian Sentiment Analysis) offers several advantages over comparable datasets in the Persian language. Notably, this dataset consists of 253, 368 comments, each categorized into 5 classes. LDPSA represents the sole extensive Iranian dataset suitable for scrutinizing educational content and e-books. Moreover, significant insights were gleaned from data analysis. For example, during the COVID-19 pandemic, Iranian individuals dedicated more time to studying and engaging with educational platforms significantly. Nearly 80% of users expressed favorable opinions concerning the informational materials available on the Fidibo website. Users' inclination towards utilizing audio books has escalated, along with other analysis referenced in the paper.

Criminal fraud through bank transactions is one of the main concerns in the field of financial services that can be carried out by individuals or organizations. These illegal activities are challenging to identify. Machine learning algorithms and integrating different features may be able to detect hidden patterns in data, but they are unable to define the structure generated by interactions between distinct features. On the other hand, a criminal transaction may not seem suspicious at first glance due to the complexity and variety of fraud patterns. Consequently, network theory should be used to analyze card-to-card transactions. In this paper, a meta-classifier and an ensemble approach based on graphs and hidden Markov chains (GHM) are utilized to detect suspected organized fraud practices. Based on the proposed method, network features and feature vectors are first extracted from the transaction graph. Then, a hidden Markov chain is applied due to the high dependence between the values of each feature extracted at successive times and the significance of the transaction data's inherent sequential nature in detecting fraud patterns. Finally, to discriminate between legitimate and fraudulent clients, meta-classifiers using neural networks, SVMs, Naï, ve Bayes, and K-Nearest Neighbors are used and compared. There are 162, 493 transactions in the studied dataset. According to the results of the experiments, the Naï, ve Bayes meta-classifier outperformed other approaches with the highest detection accuracy.

As is well-known, early cancer detection plays a crucial role in successful treatment. Since the early 2000s, wireless biomedical sensors have shown promise in this area by detecting cancer cells through the identification of protein-level alterations. Moreover, with the decreasing costs of genetic analysis, it is becoming increasingly feasible to envision the development of biomedical IoT devices shortly. These devices would be capable of real-time analysis of biological samples (blood, urine, etc. ) and, by taking into account an individual's medical history, could assess the risk of cancer presence. In this research, we demonstrate that by improving the classification speed of single-cell ATAC-seq data using machine learning methods, we can pave the way for close to real-time cancer risk assessment via the classification of samples collected by IoT devices. Among evaluating six well-known methods for classifying these data, we show that our proposed method, can achieve similar classification accuracy to the state-of-the-art single-cell ATAC-seq data analysis methods, while requiring only about a quarter of the processing time. The proposed method can provide an efficient method for rapid cancer monitoring on Internet of Biomedical Things.

In this article, the face detection and conformation of its salient features from noisy and blurred photos employing convolutional neural networks on the web are perused. Face detection has many applications primarily on the Internet, for instance, in web-based marketing data analysis, user interactions, website analytics, Internet security, entertainment and gaming such as usual graphics, various applications, the interaction between robots and their environments, video editing, legal issues, and, of course, it is not limited to these cases. Among the difficulties in this issue, we can mention the existence of intra-personal variations, rotation, occlusion, the expensiveness of the joint datasets, and the adaptation of the individual's facial expression with his face expression in the probe set. Because of the presence of delicate edges in face images, this paper eliminates blur from face images by exploiting convolutional neural networks,then, the face is detected by the conformation of its salient features using these networks. Experiments have been performed on the FDDB database and WIDER FACE benchmark.

An important development can be observed in the current environment of rapid technological progress in the healthcare industry, which can be demonstrated by the emergence of telemedicine as a key paradigm. Extensive use of telemedical techniques has happened with a particular focus on Tele-Pneumonia examinations, which provide doctors with crucial resources in remote diagnostic situations. In response to the necessity of accurate pneumonia prediction, we present a novel architecture PneuTeleCNN that makes utilization of pre-trained models such as DenseNet-121, Visual Geometry Group (VGG-16), MobileNet, and InceptionV3. This thorough approach of PneuTeleCNN meticulously evaluates how effectively the system detects pneumonia in web-based telemedicine environments, establishing the stage for improved remote healthcare solutions. These model’, s performance is rigorously evaluated using key performance indicators, including accuracy, F1-score, recall, and precision. In addition, the evaluation includes sophisticated measures like the confusion matrix and receiver operating characteristic (ROC) curve, which offer an intricate perspective on the model’, s performance. To help in the identification of the model that performs better in this scenario, a comparative study is made more more straightforward by presenting of each model’, s validation accuracy and validation loss graphs. This PneuTeleCNN framework underscores the commitment to ensuring the robustness and reliability of the proposed telemedical approach for pneumonia detection in remote healthcare scenarios.

In computer vision applications, corners are often regarded as desirable features due to their simplicity and low coordination requirements. Traditional intensity-based algorithms identify corners by examining the intensity relationship between neighboring and local regions, as well as the derivative information. Most detectors that solely utilize intensity information were developed before 2000, with FAST being an exception. Our approach is a new intensity-based corner detector that stands out by relying solely on pixel intensity for corner detection. We accomplish this by employing an innovative corner response function. Our method identifies corner locations by solely considering intensity values within a 3×3 neighborhood. By sorting pixels based on intensity and calculating the difference between one-third of the largest and smallest values, we generate a highly effective corner response map with strong discriminatory capabilities. Experimental evaluation on benchmark images demonstrates the superiority of our detector compared to seven established methods. Our method achieves better accuracy in corner localization and reduces both missed corner detections and false positives. Also, it requires only one parameter for adjustment, making it computationally efficient and allows for real-time processing potential. Furthermore, the generated corner response map holds promise for integration with deep learning architectures, opening possibilities for further exploration.

Energy plays an important role in supporting modern life, and the need for efficient energy solutions is more important than ever. This research paper explores the unprecedented potential of AI-based AI to transform the energy industry through the use of smart solutions. Using the most advanced data science, AI algorithms and machine learning models, we can unlock valuable insights, improve energy efficiency models, control energy distribution and reduce costs. This article provides an in-depth look at the various applications of artificial intelligence in the energy industry and covers the development of artificial intelligence in energy integration, demand planning, strategic planning, security and energy management.

Developing a distinct brand personality enables companies to differentiate themselves from competitors and effectively engage with customers. However, evaluating the customers’,brand personality perceptions is a challenge, as traditional methods are costly and not objective. In this study, we focus on the brand personality dimension of “, competence”,and leverage the social networks to analyze the perceptions of Persian customers. For this purpose, the comments written in the Persian language regarding the brands or users’,experiences with the brands are extracted from social networks. Then, the natural language processing techniques such as TF-IDF and word2vec are employed to prepare the data for developing machine learning models. These models categorize users' comments into three classes: aligned, non-aligned, and neutral. The classification depends on whether the comment is in line with the brand's competence, against it, or neutra. The k-nearest neighbors, Naï, ve Bayes, Artificial Neural Network and long short-term memory (LSTM) are trained on the dataset. The results demonstrate that the LSTM model surpasses the performance of other models by achieving the f1-score of 93 percent. Finally, the LSTM model used to evaluate the customers' perception of Snowa's brand personality, as a case study.

This paper explores the academic discourse surrounding cryptocurrencies, a digital or virtual currency that uses blockchain technology. The study uses bibliometric analysis to understand research trends, geographical/country contributions, institutional and departmental analyses, keyword occurrences and data visualizations to identify central topics and themes in the literature. The dataset of last five years was retrieved from the Web of Science database on February 12, 2024 and analysed. The results showed that China is the most significant contributor, followed by the USA and England. The University of London had the highest number of articles, followed by Ho Chi Minh City University of Economics and Lebanese American University. The study also identified two key topics: "cryptocurrency" and "bitcoin. " The research was categorized into five clusters based on economic, financial, and technological trends. The analysis identified influential papers in cryptocurrency research, such as "A survey on the security of blockchain system" and "An overview of smart contracts: Challenges, advances and platforms. " These papers highlight a growing interest in security concerns and cryptocurrency as an alternative investment platform, particularly during times of crisis.

Lung diseases are one of the most important areas in medicine that significantly impact individuals' quality of life. Accurate diagnosis and appropriate timing of these diseases are crucial. In recent years, with the advancement of technology and the utilization of deep learning methods and neural networks, the diagnosis of lung diseases is performed relatively quickly and accurately. Utilizing the best technologies and algorithms can be effective in diagnosis and treatment of lung diseases. Therefore, investigating, comparing, and selecting the best algorithms for the development and advancement of future work is essential. Advanced technologies in online software and web-based platforms, along with medical knowledge, play a significant role in enhancing the diagnosis and treatment of lung diseases. This process not only aids lung health but also plays a fundamental role in prevention and treatment. This study introduces prominent and new architectures of machine and deep learning-based in lung disease diagnosis using radiographic images. It also examines the role of these architectures in increasing the accuracy, speed of lung disease diagnosis and presents advanced solutions for improving the process of lung disease diagnosis on the web. The obtained results suggest that deep learning methods perform significantly better than traditional feature-based classifiers.

Diabetes is a condition related to metabolism that arises due to various reasons. Each year, a significant portion of patients experience life-threatening complications leading to fatality. This illness is classified into three distinct forms: Type 1, Type 2, and Gestational diabetes. It is of utmost importance to predict type 2 diabetes, which arises from cellular insulin utilization deficiency or secretion disorder, as it enables the prevention of complications or delays in the onset of the disease. Predicting the occurrence of illnesses using machine learning and artificial intelligence can substantially mitigate its costs. However, due to interpretability challenges in proposed models, physicians and patients are hesitant to adopt them. Previous studies have utilized various algorithms including Naï, ve Bayes, SVM, KNN, and decision tree algorithms for patients’,classification. In this paper, conducted on the Pima dataset, we employed a preprocessing method utilizing the most significant features selected by the Random Forest algorithm. Additionally, for model testing, we utilized the SVM algorithm, known for its high discriminative power in binary classification tasks and its relatively good interpretability. The results indicate that the proposed model achieved an accuracy of 80. 09%, outperforming other models by a 2. 26% improvement. Furthermore, there were notable improvements in precision and specificity metrics with the proposed model. By utilizing these methods, web-based applications can be employed to motivate physicians and patients for diabetes prediction.

Given the increasing reliance of critical infrastructure on information and communication technology, the timely detection and prevention of attacks have become paramount. Extensive research in field of neural networks and deep learning being used due to the being compatible on large datasets has been devoted to this area. Previous studies have shown that combining neural network algorithms, particularly the Convolutional Neural Network and long short-term memory, significantly improve attack prediction compared to either CNN or LSTM models individually. This study introduces a novel parallel model that integrates these two networks. The parallel networks receive two inputs simultaneously, one for sequential processing by the CNN neural network and the other for processing by the LSTM network. Each model processes the data independently, and their outputs are merged to produce the final result. The integration of CNN and LSTM models in parallel, which extract unique features and temporal characteristics from input data through convolutional and recursive layers at the same time, achieved higher accuracy than previous studies. By utilizing the well-known NSL_KDD dataset, the proposed model in this study achieved an accuracy of 99. 45% in detecting Denial of Service attacks, surpassing previous studies on the same dataset that achieved a maximum accuracy of 99. 20%.

Based on the most recent statistics, breast carcinoma stands as the most widespread cancer globally, claiming the lives of nearly 900, 000 individuals annually. Detecting this disease at an early stage and providing an accurate diagnosis can enhance the likelihood of favorable outcomes, subsequently reducing the mortality rate. Early diagnosis indeed plays a role in hindering its proliferation and safeguards individuals from falling victim prematurely. When researchers delve into the differentiation of benign and malignant tumors, and when they seek to draw conclusions about early and advanced stages of breast cancer, they confront a myriad of challenges. The limited availability of training data samples has been recognized as a fundamental challenge in this field. Due to the novelty of this research area, large datasets for breast cancer detection using thermography are not yet available, which necessitates the use of supplementary methods to compensate for the data scarcity. In this paper, to enhance the model and address the data scarcity challenge in breast cancer detection from thermographic images, we employed a combination of transfer learning and the application of Generative Adversarial Network (GAN) neural networks on the well-known DMR-IR dataset. The performance of neural networks, with and without this combination, was tested and compared. The results demonstrate that the utilization of the ResNet-152 transfer learning model on this dataset achieved an Accuracy of 83% in breast cancer detection. Furthermore, after applying the GAN neural network, the Accuracy of this same scenario increased to 90%.

Nowadays, the simultaneous consideration of spatial features and temporal features in the form of spatio-temporal data has been considered in various applications. Among the most important of these applications, we can mention the classification of drivers, the personality of users, the prediction and monitoring of the route and many other cases. Due to the lack of labels for this type of data, unsupervised methods are the only analytical methods on the table. Today, machine learning or deep learning algorithms are very popular and widely used due to the many successes they have had in solving various problems, but due to the black box nature of these methods, the internal mechanism of these models is unclear to users. On the other hand, in addition to the high accuracy of the model, the interpretability and root-finding of analyzes based on spatial-temporal data are very important. In this paper, we intend to cluster spatio-temporal data by focusing on the interpretability of results and selected interpretable features.

By the emergence of digital communication tools in human life such as smartphones and their apps, we have observed a new revolution in lifestyle. In this situation, hypercompetitive markets are established between designers and developers of mobile apps. Attracting users to download and use mobile apps is a critical issue for designers and developers of such apps, however, the factors affecting the user decision to download an app are scattered in previous studies but this study try to provide a comprehensive approach by attention to different aspects of mobile app design with emphasize on aesthetic variables. The purpose of this paper is to provide a comprehensive approach to factors affecting mobile apps downloads by emphasizing on the role of aesthetical components of an app icon. The developed research models try to test all important factors in this regards. The research population consists of game apps at Cafebazaar Play Store in Iran. By using the combined sampling method (both quota and judgmental), the relevant data of 384 apps were collected. MATLAB software package was used to evaluate aesthetical components and, then, collected data was analyzed by stepwise regression test in SPSS software package. The results showed that among identified factors, an app average score, image symmetry, designing style, anthropomorphism, icon complexity, application language and payment methods, all have a significant impact on downloading an app by users.

Routing within the Internet of Things(IoT) networks relies on utilizing the Routing Protocol for Low-Power and Lossy Networks (RPL). The RPL protocol is equipped on the sensor nodes, which can be exposed to different insider or outsider attacks because of the non-resistant nature of these nodes and their limited resources. The security issues of the RPL protocol may restrict its ability to be adopted globally. Therefore, a comprehensive analysis of RPL-associated security attacks and their consequences on the underlying network is required. This paper examines how the RPL network is affected by decreased rank attack in both moving and stationary network settings. The results highlight the destructive impact of this attack on the routing structure, leading to decreased PDR and throughput, increased average delay, and control overhead. Furthermore, our findings emphasize the provision of security mechanisms for IoT networks by highlighting the high impact of the attack in high-loss mobile networks.

The fake news story refers to information falsified or made up to manipulate or deceive the public. It is a serious issue that undermines the credibility of information sources and influences public opinion, discourse, and important decisions and outcomes. The content and source of news articles and stories must be considered when identifying and classifying fake news. Detecting fake news is an important and active research area with many potential applications and implications for society. Many challenges are involved in determining whether a news article is authentic. Several deep learning models that employ natural language processing (NLP) have shown excellent results in detecting fake news. To assess the truthfulness of news articles, our methodology is based on state-of-the-art language models based on transformers. Bidirectional Encoder Representation from Transformers (BERT) and Robustly Optimized BERT Technique (RoBERTa) is one of the most advanced models. Our findings reveal that the BERT model achieved an accuracy of 64%, while the RoBERTa model slightly outperformed it with an accuracy of 66%. These results are particularly significant when compared to similar research in this domain, which reported a maximum accuracy of 62% for both models on Liar dataset.

As long as a computer system is connected to the Internet, it is susceptible to attack as a victim. In computer networks, it becomes important to manage the network based on parameters such as network size and network data. Firewalls are devices that help network administrators in this case to establish security in the network, and can be based on the rules that the firewall is based on. It is configured to control incoming and outgoing network traffic. Firewalls can be considered the most vital components of the network in establishing security. Firat University introduced a dataset containing firewall logs with multiple classes for firewall decisions. This study uses data mining techniques to improve the validation performance of classification using various machine learning algorithms like neural networks, deep learning, and kNN. The experimental results show more than 10% improvement according to precision and recall rates among various folding scenarios used in related works with minor improvement in accuracy, too. The decision tree algorithm is fast and explainable versus other algorithms.

Infections during the neonatal period are one of the most critical factors leading to mortality in neonates in the neonatal intensive care unit (NICU) within the first 28 days of life. The majority of hospitalized neonates in NICU are premature and highly susceptible to nosocomial infections due to their compromised immune systems. Therefore, the objective of this research is to develop a model to predict neonatal infections, aiding in the early detection and management of infections among vulnerable neonates. The study involves neonates hospitalized in the NICU, with data collected from 113, 378 neonates admitted in the year 2022. Initial features for creating predictive models of neonatal infections were obtained by examining relevant sources of information and consulting with physicians and relevant specialists. In this research, data mining classification algorithms were used to create predictive models for neonatal infections. To evaluate the created models, the Recall, Accuracy, Precision and F1-Score indicators were utilized. Among the methods used, the Random Forest algorithm demonstrated the best performance in predicting neonatal infections. Among the four methods employed for balancing the data, the folding method notably improved the performance of models. Additionally, using a dataset that includes only maternal features can significantly contribute to predicting neonatal infections before the infant's birth.

The transformation of fashion through online platforms has spurred a need for high-quality clothing search engines, facilitating seamless product discovery for global consumers. However, this transition has brought forth challenges in categorization and description standards among retailers and search engines, stemming from the inherent complexity and variability of fashion items. To address these challenges, deep learning techniques like Multiple Convolutional Neural Networks (MCNNs) have gained prominence in the fashion industry. MCNNs offer autonomous learning and enhanced feature extraction capabilities, proving successful in various classification tasks, including clothing classification. In our research, we implemented a multiple-CNN architecture inspired by recent architectural innovations to elevate classification accuracy while preserving computational efficiency. We attained a classification accuracy of 93. 08%, surpassing previous benchmarks. However, the integration of multiple CNNs introduces complexities such as heightened computational demands and resource requirements. To strike a balance between achieving superior accuracy and managing the inherent complexities in clothing image classification, we designed the MCNN-14 model.

We live in a world of information, and with the ever-increasing rate of content growth, we have no choice but to use machine-based solutions to manage, classify, and use it. However, the produced content is often unstructured, and for computer systems to be able to read and process it, there is no choice but to convert it to structured text. Knowledge graph is one of the ways to represent data in a structured way, and its roots are found in the semantic web. There are different approaches to creating a knowledge graph, but given the significant advances in artificial intelligence, the use of natural language processing is one of the most accurate and automatic ways to do this. In this paper, an attempt has been made to review the latest methods proposed for creating a knowledge graph based on natural language processing. This first allows us to have a better understanding of this research area,in addition, the approaches used by researchers in this field are reviewed and the pros and cons of these approaches are identified. Also, the existing research areas for further work and even the initial ideas for new creative solutions are provided for researchers.

Infants during the neonatal period, defined as the first four weeks after birth, exhibit the highest probability of mortality. This vulnerable stage of early life, characterized by heightened rates of mortality and neonatal diseases, underscores the susceptibility of neonatal life during this period. Consequently, delineating the mortality profile of neonates within the community is a pivotal strategy for identifying causal factors and presenting findings, thus constituting one of the most crucial approaches for enhancing neonatal health outcomes. In the field of medical science, one of the most prominent applications of machine learning is in disease diagnosis and prediction. Therefore, the aim of this research is to introduce two methodologies designed to forecast the likelihood of neonatal mortality. The first approach relies on utilizing machine learning classification algorithms, while the second approach employs convolutional neural networks (CNNs) on non-image data. To assess the developed models, metrics including Accuracy, Recall, Precision, and F-Score have been utilized. Among the methods used, the second approach, which uses CNN, performs better in predicting the probability of infant mortality during the neonatal period with 98% accuracy.

The purpose of this study was to investigate the factors influencing the willingness to use wearables equipped with Internet of Things technology in championship sports. The statistical population for this study consisted of Kashan athletes who achieved national and international championship titles in 2023. The research findings indicated that the variables of expected performance, effort, and social influence have a positive and direct impact on the attitude toward using wearables. Additionally, the variables of facilitating conditions and attitude demonstrated a significant influence on the willingness to use sports wearables that are equipped with Internet of Things technology. To capture the attention of professional athletes, it is recommended to launch diverse campaigns featuring influential individuals and providing high-quality content. The focus should be on raising awareness about the capabilities of these technologies and the power of sports devices to enhance the effectiveness of athletic activities, improve performance, and achieve greater success.

A facet of sentiment analysis is Aspect-based Sentiment Analysis (ABSA), which entails identifying and assessing sentiments towards particular aspects or attributes within text data. In ABSA, an induced tree can illustrate the connections between various aspects of a text and the corresponding sentiments. By employing supervised contrastive learning, models for ABSA can be trained using both labeled and unlabeled data to boost performance. In this research, we fused induced tree structures from the pre-trained BERT model with supervised contrastive learning to empower the model to effectively distinguish between different aspects and sentiments. This method assists the model in grasping subtle sentiment variations towards various aspects of products or services, leading to more accurate sentiment analysis outcomes. The experimental results showcase the effectiveness of our integrated framework in delivering precise sentiment forecasts and profound insights into sentiment-attribute relationships, as evidenced by the enhanced classification accuracy on SemEval2014 benchmarks.

The sustainability of digital human resources management in organizations requires attention to soft behavioral components as the basis of human interaction. This means respecting the human dignity of employees and fulfilling the responsibilities of managers towards internal customers and external audiences. The purpose of this research is to identifying technology-oriented infrastructure, required processes and the role of employees in digital human resources management with a practical and developmental approach. Therefore, in the first stage of the research, after studying the scientific documents and determining the framework of the qualitative research method and the techniques used in it, a semi-structured interview was conducted with the experts. The opinions collected from 14 human resource managers or digital transformation project managers in the studied companies were analyzed by thematic analysis method. The main components were identified as intelligent technology, intelligent knowledge management, organizational learning, employee participation and the development of employee intelligence. In the second stage, the fuzzy DEMATEL method was used to determine the weight of the indicators and build a causal network, in which the organizational learning component has the highest priority over other components.

Facial emotion recognition has been extensively researched in recent years due to many use cases. The most important applications are increasing human-computer interaction and helping with autism spectrum disorders. Also, in most applications, real-time execution is required. The model and computational resource are two main factors of the inference time. Hence, to propose a real-time method, it is required to concentrate on these two factors. In this paper, we utilized EfficientNetV2 due to its efficiency. Furthermore, we proposed a scalable method based on resolution scaling to keep the model in real-time in different computational resources and models. This scalable method has been implemented using a polynomial equation to find the best value of the resolution for a specific inference time based on our hardware and model. Thus, the main objective of this paper is to propose a scalable real-time method for the facial emotion recognition task using resolution scaling. Consequently, using a polynomial equation for resolution scaling, we proposed Scalable-ENV2B0 and Scalable-ENV2S based on EfficientNetV2B0 and EfficientNetV2S, respectively. According to the ultimate results on the KDEF dataset, Scalable-ENV2B0 can classify (302, 302, 3) input size images in real time on our hardware. Also, this model achieved an impressive 96% accuracy on KDEF, which outperforms previous real-time studies based on our knowledge. However, the main advantage of the proposed method is scalability, which hasn’, t been addressed in this task so far.

According to the number of users on social media platforms and the number of followers of published content, Fake news in social media could adversely affect society and governments. In this paper, a new approach to Fake news detection is presented. Our approach is to provide a hybrid model in order to incorporate different types of features each of which is extracted by the corresponding component in the model. A transformer model generates the input of each component, namely sentiment analysis, hate speech detection, and topic modeling. Thereafter, a combination of hidden layers is performed through concatenation, which is then fed into the final classifier used for fake news prediction. We evaluated the proposed model by collecting 2500 Persian data (related to COVID-19) from the social networks complemented by various preprocessing tasks. The results demonstrate a higher accuracy (87%) for the proposed approach compared to the standard baselines.

In today's modern world, the prominence of social media is undeniable, with Twitter standing out as a pivotal platform for global communication and information sharing, particularly in expressing and amplifying emotions. In this digital era, sentiment analysis has emerged as a crucial tool for measuring emotions and reactions to daily events, especially in the context of business improvement. It enables businesses to rapidly and effectively decipher trends and customer insights. Within the realm of sentiment analysis, we encounter a diverse range of models, each with its unique features and limitations. This research aims to amalgamate the strengths of various approaches by integrating the Naive Bayes classifier, a bespoke rule-based model, and BERT—, a relatively lightweight transformer model—, particularly in the context of sentiment analysis of Persian Twitter media. Our findings reveal that traditional models such as SVM, Naï, ve Bayes, and MLP alone do not yield high-quality results. Our hybrid model, when used independently, outperforms BERT, achieving an accuracy of 89% compared to BERT's 86% which represents a significant advancement in sentiment analysis. Although slightly more structurally complex, it maintains computational intensity on par with BERT fine-tuning while outperforming BERT when used individually. This advancement stems from our unique approach of integrating Naive Bayes and a bespoke rule-based model, subsequently leveraging BERT for sentiment classification, thus enhancing its effectiveness in social media contexts.

The rise of novel technologies significantly impacts the expansion, achievement, and resilience of businesses and their divisions like marketing and sales. Among these technologies, augmented reality stands out prominently. Numerous companies leverage this technology to enhance the shopping and consumer experience. But the question is, are consumers willing to use this technology in their purchasing process or not? Therefore, the current research aims to analyzing augmented reality technology acceptance models by consumers. To achieve the goal of the research, the researchers used the method of systematic literature review. After screening the articles, finally, 14 articles were left to extract the findings. The results showed that the acceptance model of augmented reality technology by consumers includes the following factors: Effort Expectancy, Perceived usefulness of technology (PU), Perceived ease-of-use (PEOU), Aesthetics, Performance Expectancy, Familiarity with AR, Customer engagement, perceived enjoyment (PE), Facilitating Conditions, Cognitive dissonance, Consumer choice confidence, Curation, Perceived fee, Reward, Attitude, Perceived value, perceived informativeness (PI), Perceived Interactivity, Satisfaction, Innovation resistance, Subjective Norms, Trust, Social influence, Social status, Education level, Monthly income, Age, Self-identity, Consumer experience, Gender, Personal innovativeness.

In the domain of Natural Language Pro-cessing (NLP), the integration of Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) represents a significant advancement towards enhancing the depth and relevance of model-generated responses. This paper introduces a novel hybrid RAG framework that synergizes the Sentence-Window and Parent-Child methodologies with an innovative re-ranking mechanism, aimed at optimizing the query response capabilities of LLMs. By leveraging exter-nal knowledge sources more effectively, the proposed method enriches LLM outputs with greater accuracy, relevance, and information fidelity. We subject our hybrid model to rigorous evaluation against bench-mark datasets and metrics, demonstrating its superior performance over existing state-of-the-art RAG tech-niques. The results highlight our method’, s enhanced ability to generate responses that are not only contex-tually appropriate but also demonstrate a high degree of faithfulness to the source material, thereby setting a new standard for query response enhancement in LLMs. Our study underscores the potential of hybrid RAG models in refining the interaction between LLMs and external knowledge, paving the way for future research in the field of NLP.

The increasing expansion of mobile phones along with the expansion of the possibilities of these phones has provided a suitable field for information theft. Android is undoubtedly the most popular and widespread operating system of mobile phones, which has become the target audience of many malware authors due to this expansion. This article seeks to provide a suitable and powerful solution for detecting malware. Data processing uses a combined feature selection operation. This idea extracts the most important features and improves the accuracy and speed of detection. Then, three-level stacking is used for the detection stage. This method can significantly improve the accuracy and power of generalization compared to other methods based on the innovative idea of dataset separation. The accuracy of this method is equal to 99. 5.

Melanoma is one of the types of cancerous skin lesions, where early detection is crucial to prevent patient mortality. One method for early detection of melanoma involves using dermoscopic images of skin lesions to train deep learning models, which can then be used to classify skin lesions in patients, including the diagnosis of melanoma. A significant limitation of deep learning models is their need for substantial amounts of labeled data. This article discusses data augmentation using the Wasserstein GAN (WGAN) network to overcome the issue of limited diversity in images generated by GAN networks (a problem known as Mode Collapse). By generating 5, 000 high-quality synthetic images of the melanoma class and adding these images to the unbalanced HAM10000 dataset, an improved accuracy in diagnosing this disease was achieved using the pre-trained deep ResNet50 model. The proposed model improved melanoma classification accuracy by 10% without significantly altering the overall model accuracy. These results suggest that using the WGAN network for data augmentation can enhance the classification accuracy of melanoma.

Social networks empower individuals to freely share their perspectives on a diverse array of subjects. One such topic is the impact of the coronavirus vaccine in preventing the disease. People have written their varying reviews on this matter through tweets. These reviews would contribute to grasp people’, s feelings and sentiments about Covid-19 vaccination. One common method employed by businesses to assess sentiment in social data is Sentiment analysis. Our model classifies individuals' perspectives into three labeled data categories: negative, neutral and positive. We have used Few Shot Learning to offer a cost-effective training method, particularly paraphrase-mpnet-base-v2, to improve text classification of the Kaggle-extracted tweet dataset. The findings derived from the experiment suggest that our method has achieved 96. 37% accuracy score, , which outperformed previous published works.

Android devices are providing about 70% of the web traffic. Therefore, the security of the android devices is one of the major factors impacting the web security. Autonomous detection of the malware infecting Android devices using machine learning methods can act as a scalable solution for security provision on smartphones. This study aims to introduce an innovative approach for detecting mobile phone malware by leveraging users' emotional reactions and interactions with their devices during sudden and unpredictable events. Traditional mobile malware detection methods that rely on permissions and API calls have extensively been researched, yet they often overlook human elements such as emotions and their potential implications in this context. The methodology proposed in this research involves capturing users' reactive behaviors to unexpected events using Natural Language Processing (NLP), analyzing their interactive patterns with mobile phones through clustering techniques, and employing machine learning algorithms and classification methods for malware detection. The experimental results show that the proposed method can provide an accuracy of more than 96% which provides an efficient tool for Android and web security.

Genre-based movie classification is a significant topic in text processing and machine learning. An automatic system that classifies movies by genre can help people find their top picks from an available variety. This research seeks to establish a model for genre-based movie classification utilizing machine learning algorithms and a movie dataset including details like movie titles, plot synopsis, actors, and genre information. Various methods such as logistic regression, decision tree, Naï, ve Bayes, and convolutional neural networks, combined with LSTM in Python, are used to train the movie classifier model. Using language models such as word2vec and TF-IDF, which transform words into numerical vectors, is one popular technique. This improves the accuracy of the model's genre prediction. The proposed model achieves a high F1-Score of 0. 84, which is comparable with the state-of-the-art algorithms that achieved F1-Scores of 0. 82-0. 92. Users can utilize this system to locate movies of their preferred genre by inputting the title or description and receiving recommendations. Using a web API, this system may also automatically suggest movies collaboratively.

Hallucination in large language models refers to outputs that appear correct but contradict reality or diverge from the source. Detecting hallucination in large language models is crucial to prevent the dissemination of these hallucinations in applications directly or indirectly related to such models. In this study, we have employed a simple algorithm to detect hallucination in a large language model. Our hypothesis is based on the hypothesis that if a large language model responds to the paraphrases of a question and an inconsistency is discovered among its answers, then we say that it is hallucination, and if the answers are consistent, it likely provides a correct answer. We have checked and confirmed these two hypotheses with experiments. In this way, our proposed method to discover the hallucination in answering a question is to create different paraphrases of that question and check the existence of inconsistencies or contradictions in the answers given to the generated questions. The presence or absence of inconsistency confirms the presence or absence of hallucinations. Experiments show that this method is able to detect hallucination in answering questions with high accuracy.

Answering complex questions based on multi-hop reasoning on textual data requires retrieving different documents. The main challenge in answering complex questions is the existence of few lexical relations between documents or paragraphs containing answers and questions. This Paper introduces an iterative retrieval method based on text processing techniques, which uses the structure of links inside Wikipedia while recognizing the type of question. This method allows retrieving documents related to open domain multi-step questions on the Wikipedia encyclopedia. The proposed method uses linguistic models based on a deep neural network to recognize the type of question, extract entities and keywords, and finally retrieve and rank documents. The evaluation results of the proposed method on the HotpotQA dataset show an improvement in the performance of the proposed method compared to the basic methods.

Depression is a common and serious mental illness that affects millions of people worldwide. Early diagnosis and treatment of depression are essential to improve the quality of life of those affected. Traditional methods of diagnosing depression include interviews with psychiatrists and completion of psychological questionnaires. These methods are time-consuming and costly, and some people may find it difficult to verbally discuss their depression symptoms. In this study, a Persian chatbot based on deep learning was designed and implemented to diagnose depression. The chatbot was trained using textual data related to individuals with depression and healthy individuals. The data used in this study was collected from Moshaver Co Forum, which includes question and answer exchanges and information sharing among Iranian users on this web platform, in two categories of depression and non-depression. The results show that the proposed chatbot is able to diagnose depression with an accuracy of over 85% and an F1 score of 80. 5%, which shows better performance compared to similar researches. This Persian chatbot can serve as a useful tool for early diagnosis and treatment of depression.

Efficient management of computing resources has always been a significant concern for users. An automatic scaling system can help in managing hardware resources by adapting to the system's performance history. It can increase or decrease resources automatically, without human intervention, based on predefined criteria. This ensures smooth program execution without any disruption caused by changes in the operating environment. This study focuses on serverless environments, which rely on functions. We model these functions using graph theory, analyze their dependencies, and identify the most critical bottlenecks in the graph. We then use two approaches, supervised and unsupervised, to predict the scalability of bottleneck resources. To be more sure of the scaling decision, the consensus mechanism compares the predictions of the models, and the best model's result is considered the final scaling decision, which creates consistency between the results obtained from the methods. Results show that supervised approaches perform better than unsupervised approaches in the automatic scaling problem. The models implemented in this research can determine the scaling result with 98% accuracy, which is a 2. 5% improvement compared to previous works.

In social networks groups play a crucial role and making decisions based on majority consensus. Which influencer nodes should we select if our goal is to broadcast a subject in a target group and increase the number of active nodes in this group? Here, we study a new influence maximization (IM) problem that focuses on individuals in a target group who are activated by some relevant topic or information. Target Group Influence Maximization (TGIM) aims to select k influencer nodes in such a way that the number of activated nodes in the target group is maximized. In this paper, we study TGIM and focus on activating the majority of nodes in the target group. We propose an algorithm named Reinforcement Learning for Target Group (RLTG) based on the analysis of the influence of nodes on the target group. The algorithm uses the reinforcement learning approach to learn the optimal path from each target node to some candidate influencers. The experimental results indicate that the recommended approach outperforms known methods.

Today, one of the most common uses of the Internet is searching the web and retrieving information from it. We all use general search engines like Google and Bing to search for information on a daily basis. Web crawlers are the most important part of a search engine that crawls the entire web content and extracts the content by following the links on the web pages. Focused web crawlers are a type of web crawlers that limit the crawling process to a specific section of online content and are used in vertical search engines. For example, they may only retrieve certain types of media (such as PowerPoint files). In this paper, a systematic mapping study has been conducted and the approaches used in the development of focused web crawlers have been reviewed and the advantages and disadvantages of each have been discussed. Also, 2 new approaches have been identified and introduced. This study shows that the approach based on "ontology or semantics" is the most used in the development of focused web crawlers. Also, the decision to use each of the introduced approaches depends on the available resources and the existing limitations for development.

One of the major challenges in the world today is traffic accidents, which occur due to various factors such as driver distraction during driving, as these factors will have serious and irreparable consequences. There are various methods for detecting distraction factors. The use of machine vision-based technologies and deep learning for automatic detection of these factors can improve driving safety and minimize accidents caused by distraction. The use of intelligent image processing algorithms on image data obtained from cameras installed in the driving environment for extracting relevant characteristics and signs is very efficient. By utilizing deep learning algorithms, the driver's condition can be identified and recognized. In this way, intelligent assistance is provided to the driver to improve their focus and prevent accidents. The main goal of this article is to investigate the methods proposed in the field of automatic detection of driver distraction factors using machine vision techniques and deep learning. Additionally, novel web-based methods can be utilized in driver distraction detection systems with the aim of improving safety and preventing traffic accidents.

Today, we are in the era of big data, and high-dimensional data is growing in the education system, and this rapid growth has created challenges in efficient and effective data management. Proper data management allows us to obtain the necessary knowledge from big data more quickly and accurately with methods such as data mining and machine learning. One of the ways to increase prediction accuracy in machine learning algorithms is feature engineering. Feature engineering is one of the most important steps to increase the model's predictive performance and produce a quality dataset. Many studies have shown that performing feature engineering before data classification is necessary and necessary in order to obtain optimal results. Feature selection methods as part of feature engineering increase the efficiency of the learning process. The goal of a feature selection method is to identify relevant features and remove irrelevant features in order to obtain a suitable subset of features. So as to increase the accuracy of performance prediction and this selected set is able to describe the original data set well. In this research, using feature selection methods, the effect of features on the results of predicting the performance of learners has been investigated.