INTERNATIONAL CONFERENCE ON WEB RESEARCH | Year:2023 | دوره:9 |Papers Count:87

Always variation in risk levels may alter the behavior of individual investors. Thus, give more accuracy of future variance prediction in financial time series using past data has been interest to many researchers. GARCH family models as the most common statistical methods are considered in conditional heteroskedasticity prediction that used past observations. Also, recent research shows that the observed value of a time series is relevant to other time series values. This paper in the first, use the GARCH as a statistical model to modeling and prediction the conditional heteroskedasticity financial time series, then using a genetic algorithm to find best inputs for cross relations volatility with Neural Network modeling. In order to demonstrate the performance of the proposed method compared with statistical and neural network methods, the G7 time series dataset is used. The results show a reduction of 40% in the Mean Square Error (MSE) in S&P 500 time series conditional heteroskedasticity prediction.

There is no doubt that social networks have gained prominence among users in today's society. Having received special attention from users, social network marketing has attracted the attention of businesses and individuals. As a consequence of the competition in these spaces, businesses and individuals have used some incorrect methods in order to falsely increase the engagement rate of their websites. The use of bots has been one of the most common methods of falsely increasing engagement rates. In order to discover these fake robots, various researches have been conducted. The objective of this article is to provide a review of previous research in the field of detecting robots in social networks. Based on the results of the review, it was found that random forest, nearest neighbor, logistic regression, and linear regression algorithms are more widely used than other traditional machine learning approaches. As well, convolutional neural network algorithms, long-term memory, and transformer learning have become more popular in deep learning approaches.

The Internet of Things (IoT) is a concept by which objects find identity and can communicate with each other in a network. One of the applications of the IoT is in the field of medicine, which is called the Internet of Medical Things (IoMT). Acute Lymphocytic Leukemia (ALL) is a type of cancer categorized as a hematic disease. It usually begins in the bone marrow due to the overproduction of immature White Blood Cells (WBCs or leukocytes). Since it has a high rate of spread to other body organs, it is a fatal disease if not diagnosed and treated early. Therefore, for identifying cancerous (ALL) cells in medical diagnostic laboratories, blood, as well as bone marrow smears, are taken by pathologists. However, manual examinations face limitations due to human error risk and time-consuming procedures. So, to tackle the mentioned issues, methods based on Artificial Intelligence (AI), capable of identifying cancer from non-cancer tissue, seem vital. Deep Neural Networks (DNNs) are the most efficient machine learning (ML) methods. These techniques employ multiple layers to extract higher-level features from the raw input. In this paper, a Convolutional Neural Network (CNN) is applied along with a new type of classifier, Higher Order Singular Value Decomposition (HOSVD), to categorize ALL and normal (healthy) cells from microscopic blood images. We employed the model on IoMT structure to identify leukemia quickly and safely. With the help of this new leukemia classification framework, patients and clinicians can have real-time communication. The model was implemented on the Acute Lymphoblastic Leukemia Image Database (ALL-IDB2) and achieved an average accuracy of %98. 88 in the test step.

Due to the increasing number of customers, the number of private and public banks, traditional and technological services and facilities, the need for ontology in this area seems necessary. The construction of ontology makes the transparency of customer and service communications transparent and paves the way for the use of up-to-date technologies such as machine learning and artificial intelligence in banks. The common organizational structure of the country's banks, let's present the first ontology of the banking sector in the country. In this regard, banking documents and services have been extracted by reviewing the websites of domestic banks. Also, interviews with 3 experts in the field of banking and technology and the extracted classes have been modified and finalized. It should be noted that we will use the Power Designer tool to display classes and data, and to display the graphic and content between them using Portege software. The created model includes 7 main classes and 74 subclasses.

Emerging applications (e. g., Internet of Things) in new generation wireless networks have strict information freshness requirements. Recently, age of information (AoI) has been introduced as a new QoS metric which differs from conventional measures such as delay and throughput. AoI is defined as the elapsed time since the generation of the last received packet in the destination. Optimal configuration of the transport and physical layer protocols is key to AoI minimization. In this paper, we study the problem of scheduling multi-path TCP (MPTCP) sub-flows over a dual-connectivity-based physical transmission medium based on LTE and mmWave technologies. The synergy of a multi-path transport layer and a multi-connectivity-based physical layer gives rise to an efficient communication setup for AoI minimization. In order to optimize the scheduling of traffic sub-flows over the two available paths, we propose a model-free optimization algorithm using reinforcement learning. We aim at minimizing the long-run mean AoI for the data packets received by the destination. Simulation results showcase the superiority of the proposed algorithm in comparison with the default MPTCP scheduling algorithms, i. e., round-robin and lowest RTT.

Scientists around the world study data mining extensively, but many methods are limited to analyzing small databases. Technological advances have led to the emergence of Incremental Machine Learning and Stream Data Classification to handle large amounts of diverse data. The challenge is to quickly extract information from incoming sequences of data, but the high speed and complexity of the input data limit the application of previously proposed methods. The Hoeffding tree algorithm is crucial for Stream Data Classification and employs the Hoeffding bound to select a splitting feature. In this paper, we propose a method that combines an Incremental Decision Tree called the Hoeffding tree with Ensemble machine learning using bagging to enhance accuracy. Our implementation and analysis show that our proposed method improves accuracy compared to the simple Hoeffding tree. We also analyze the algorithm with different numbers of base models and examine graph diagrams to illustrate the improvement in accuracy.

Data augmentation is a method to efficiently use the existing data to train deep neural networks. Maintaining requirement traceability links helps to improve software quality and prevent defects by aiding software development management. To ease this maintenance, automatic link recovery techniques can be used. One of the recent techniques to do this is to use a language model. We propose three code data augmentation techniques to improve language models’,performance in requirement to code traceability link recovery. These three techniques are rename variable, swap operands, and swap statements. These are general techniques that can be implemented for different programming languages, and have the capacity to generate a variety of outputs randomly, which can improve the generalization of the model. The results of the evaluations show that code data augmentation improves the language model's performance in recovering doc-method links that are similar to requirement-method links. Using code data augmentation, the precision is increased from 0. 669 to 0. 722, the recall is increased from 0. 574 to 0. 601, and the Wilcoxon test shows that the improvements are significant.

Today, the Internet is a major part of society. Given the ubiquity of the Internet, its availability is a must. Attackers, on the other hand, seek to make Internet services inaccessible and exploit Internet service companies. Attackers use various tools and methods to attack the networks and infrastructure of service companies. These attacks are also called network traffic anomalies. In general, malfunctions or attacks are network events that deviate from normal expected behavior and are suspicious of security. In general, anomalies or attacks are network events that deviate from expected normal behavior and are suspicious from a security point of view. Many different methods have been proposed to detect attacks in the network. One of the most important challenges of the previous methods is the low accuracy and lack of interpretability. In this paper, we tried to use a combination of basic methods to detect attacks and achieve 89% attack detection accuracy in the balanced dataset. This accuracy has increased by 3% compared to previous works. In order to solve the challenge of interpretability, we applied SHAP, LIME and decision tree methods and identified the effective features in detecting attacks. The proposed method, in addition to high accuracy and interpretability, has a higher speed than previous works.

Elderly people are an important age group in society that must be constantly under control. Because of old age, these people suffer from various diseases such as blood sugar, blood pressure, and obesity. If the elderly are not treated and their diseases are not controlled by taking medicine, it will cause headaches, dizziness, and blurred vision, which will cause problems for them as a result of these symptoms. One of the problems they cause is physical and mental injury. Physical injuries in the elderly population are caused by falls. Therefore, providing methods to prevent them from falling can play an important role in their physical and mental health as well as in improving their quality of life. Machine learning algorithms can predict the condition of the elderly at any moment. If an elderly person has acute symptoms, the risk of falling can be prevented by turning on the smart sensors. In this study, we propose a method for predicting falls in elderly individuals with diabetes using a multilayer perception classifier. According to these results, elderly people with diabetes are more prone to falls because of symptoms caused by blood sugar, such as dizziness and motor nerve disorders. By designing an intelligent system that can predict the condition of the elderly based on various diseases, it is possible to prevent them from falling. In this study, various factors collected from this dataset were examined. By reviewing the data and confirming the opinion of a cardiovascular disease specialist, the elderly with diabetes had more symptoms, such as headache, dizziness, and blurred vision, than the rest of the elderly,therefore, these symptoms lead to a drop in blood pressure and fall. The results obtained from the implementation of the machine learning algorithm showed that machine learning could predict the risk of falling in elderly patients with diabetes with a high performance (accuracy=97%).

In the evening where ranking on the search engine result pages is directly related to increasing the number of visitors and the progress and development of a business, search engine optimization or SEO is a process that helps to gain a higher ranking. Websites can be classified with the help of machine learning techniques based on the quality of setting SEO guidelines. Classification algorithms are combined with each other with the aim of increasing classification accuracy and are used as an ensemble classification model. In this article, we implement an ensemble classification model with the help of a random forest algorithm, which places web pages in one of the predefined classes based on SEO quality. The obtained results show that the accuracy of the constructed model is between 70. 50% and 73. 17% and is more accurate than previous works in which ensemble classification algorithms were not used. The built model can help developers build automatic software for detecting the SEO quality of web pages.

Nowadays, we are faced with a huge amount of private data generated in different ecosystems, including the Internet of Things, social networks, peer-to-peer networks, and e-commerce, to mention a few. The performance of traditional object storage and retrieval systems concerning response time is going to be degraded by a drastic increase in generated data. This problem becomes more challenging when a specific private object should be found. To address the mentioned problem, this paper suggests scalable object storage and retrieval. To do this, three well-known methods, content-addressable networks from peer-to-peer systems, vector space model from information retrieval systems, and siamese neural networks from neural networks theory, are involved in a framework to bring a scalable system to solve the mentioned problems. The suggested framework is unique to the best of our knowledge because there is no fusion among mentioned fields reported so far. A case study on signature detection is also conducted to evaluate the proposed framework. The results show that, compared to a centralized system, the proposed framework significantly decreases the response time for detecting a signature while maintaining the same accuracy. In addition, the proposed mechanism does not require the private objects to be sent to a central entity, which helps to alleviate privacy concerns.

Research on search engines has demonstrated their significant impact on information search and user behavior and highlighted the need for continued investigation of their effectiveness, efficiency, and ethical implications. Moreover, search engine optimization (SEO) has emerged as a crucial aspect of online content creation and marketing, as it involves optimizing websites and online materials to improve their visibility and ranking on search engine results pages, thereby increasing their reach and impact. SEO metrics provide valuable insight into a website's search engine performance. However, the relative importance of these metrics remains unclear as search engine algorithms and keyword competitiveness can vary widely. Therefore, ongoing research is needed to better understand the relationship between SEO metrics and search engine rankings. Given the challenge of determining the relative importance of SEO metrics using traditional statistical methods, interested in applying machine learning techniques. By training models with custom datasets of three search engine results (Google, Bing, and DuckDuckGo) and their associated SEO metrics, we hope to gain a better understanding of the complex relationships between these variables. The first two pages of each search engine return the most relevant results. In this article, we propose two types of methods to do this. First, classify the page index of web search results. Second, learning ranking methods to rank them. LambdaRank has the best NDCG compared to other methods for Google and DuckDuckGo which are 0. 86 and 0. 93 respectively. RankNet is the best learning-to-rank method for Bing with 0. 85 as the NDCG value. Logistic regression has the highest accuracy compared to other methods in page index classification for Bing and Google, which are 61. 29% and 64. 71%, respectively. K-nearest neighbor performs best for DuckDuckGo with an accuracy of 66. 67%.

Identifying, training, and hiring new workforce is one of the greatest challenges for any organization. Hence, managers of organizations tend to keep their professional workforce for a long time. This issue is more critical for the human resource managers of banks because it is difficult to attract expert personnel who are familiar with the financial and credit field. Therefore, to clarify the challenges in work environment, we look for a method to investigate this issue by exploring and analysis of opinions of banking industry employees. The method used in this research is the non-negative matrix factorization method. To perform the model optimally, two feature engineering techniques called TF-IDF and Bag of Words with two types of functions, Frobenius norm, and Kullback-Leibler divergence have been implemented. To evaluate the model, we used the silhouette clustering index and the cohesion criterion. Finally, the BoW method with the KL-Divergence function is the optimal model with the silhouette criterion with a value of 0. 36 and the cohesion criterion with a value of 0. 8945. After modeling, 12 major topics that are of interest to bank employees have been identified. After analyzing results obtained from NMF modeling “, management”, , “, career opportunity”, , “, people and organization”,are the most dominant topics that appear in employee opinions with the frequency of 12. 44%, 9. 78%, and 9. 46%, respectively. On the other hand, “, customer service”, , “, role-play on teamwork”,and “, organizational culture”,have the least impact on how employees evaluate the work environment with the frequency of 6. 22%, 6. 53%, and 6. 89%. The results of this research can provide better insight for human resource managers in Australia for evaluating banking industry employees' performance in the workplace.

In terms of death rates, breast cancer comes in second, among women with cancer. Despite the fact that cancer cells grow in a multistep process involving a number of different types of cells, prevention of breast cancer stays a challenge inside the modern world. As a method of breast cancer detection, this paper proposes ENTROPYMOC strategy, a fuzzy decision tree with a new formula of Entropy. It aims to improve the classification accuracy, precision, recall and F1-Measure of the decision tree by overcoming the limitations of the ID3 algorithm, which is not able to classify continuous-valued data. In the field of machine learning, fuzzy decision trees are becoming increasingly popular. This algorithm reduces the complexity of the logarithmic entropy formula by simplifying the Shannon entropy principle. WBCD (Original), WDBC (Diagnostic) and Coimbra datasets are used to test the improved algorithm. Based on the experimental results, the improved fuzzy-ID3 algorithm outperforms the other four classification algorithms (SVM, Naï, ve Bayes, Random forest and FId3) in terms of accuracy. In Coimbra dataset, accuracy increased by 3. 448%.

Discovering emerging entities (EEs) is the problem of finding entities before their establishment. These entities can be critical for individuals, companies, and governments. Many of these entities can be discovered on social media platforms, e. g. Twitter. These identities have been the spot of research in academy and industry in recent years. Similar to any machine learning problem, data availability is one of the major challenges in this problem. This paper proposes EEPT. That is an online clustering method able to discover EEs without any need for training on a dataset. Additionally, due to the lack of a proper evaluation metric, this paper uses a new metric to evaluate the results. The results show that EEPT is promising and finds significant entities before their establishment.

The current research was conducted with the aim of explaining the information-seeking behavior of Allameh Tabataba'i University faculty members based on Belkin Episode Model. In terms of nature, it was of practical type and was done with survey-analytical method. The research community consisted of 562 academic faculty members of Allameh Tabataba'i University, and the research sample consisted of 226 people who were selected by stratified random method. The tool used to collect data was a semi-structured questionnaire. The collected data were analyzed using spss software. The findings of the research showed that faculty members used the keywords and key phrases suggested by internal and external databases "on average",The specialized terms searched in internal and external databases have "moderately" helped to meet the information needs of faculty members,The user interface environment of the internal and external databases, in terms of the display method and type of resources, is in "moderate" agreement with the preferences of the faculty members, and the keywords and key phrases searched by the faculty members with the content of the retrieved resources, as well as the capabilities designed in The search system of internal and external databases is "moderately" compatible with advanced exploration tools. As a result, it seems that the suggested keywords and key phrases of databases, the user interface of databases and the features of the database search system should be more user-oriented to help solve the information needs of faculty members.

About 10% of insurance claims are fraudulent according to published reports. Insurance companies can take a very robust approach to detect anomalies using machine learning techniques. This study proposes a new model based on regression-based machine learning algorithms to predict the Total Price of a patient's claim based on the history of other patients, and then compare the estimated amount with the actual amount to obtain their price difference. The abnormal or fraud costs will be predicted in claims based on a threshold for the absolute price difference. A dataset of 99, 440 records of RASA web portal is gathered for evaluation. Deep learning has the best mean absolute error (MAE) in the training phase, but the decision tree has the best MAE in the testing phase. So, the decision tree is used for anomaly detection, which can detect about 17% of records as abnormal with at least a 30% deviation. Expert human assessors check the results and approve more than 50% of reported anomalies.

With the development of science and technology through social networks, the process of diffusion has become a controversial subject. In social networks, the process of diffusion is transmitted from one individual to another. The influence maximization problem, in which at least k nodes are chosen as the initial nodes of the spreader in the smallest amount of time, is one of the most critical aspects of the diffusion process. Furthermore, the initial nodes in the social network should generate the most influence on the other nodes in the network. However, in the algorithms proposed to solve the problem of influence maximization, the real-world factors which affect the influence probability in diffusion models are ignored for example, the influence probability is determined randomly. This problem causes influence maximization algorithms based on diffusion models to face the challenge of time-consuming Monte Carlo simulation. So, the RIM algorithm presented in this article which this method selects the initial nodes of the spreader according on their local reachability. In addition, Due to privacy in social networks, complete information about people such as gender, education, etc. is not available. For this reason, in the RIM algorithm, using the graph structure in social networks, the influence probability is calculated in the new DBT diffusion model. In addition, the RIM algorithm exhibits stable performance in the selection of the spreader's initial nodes in terms of influence spread, and execution time.

The estimation of the human pose is an interesting computer vision problem. The principles of fuzzy logic can be used to distinguish the human poses due to the existing uncertainty. In this paper, we suggest fuzzy logic-based modeling for identifying human actions. Fuzzy membership functions that highlight the discriminative pose connected to each action are considered for feature extraction. Additionally, to identify human activity, a multilayer perceptron classifier is applied. For a more accurate classification of yoga poses, various methods of estimating poses and identifying key points were discussed in detail. Evaluations of the proposed method on the benchmark datasets indicate the performance of the proposed method.

It is essential to analyze scientific literature when conducting review studies (systematic, narrative, etc. ). Review articles can improve in quality by choosing or incorporating papers with high research impact. The quality of research has been measured using a variety of indicators. These metrics primarily address certain characteristics like the citation index. It is impossible to study the caliber of research in any field on an individual basis. It has to do with connections. Therefore, it would be advantageous to create a network of research items. In this study, we introduce a novel tool for the analysis of metadata in scientific literature. We tested our technique on the literature of breast cancer. The tool extracted 49, 604 papers resulting in 575, 894 nodes and 1, 532, 328edges. We looked at the topological and structural characteristics of the constructed network, briefly. However, this tool can be utilized in any other domain of interest.

Denial-of-service attacks are always one of the most widespread security attacks at the enterprise network level. In DDOS attacks, a large amount of false demand is intentionally sent to the target network to disable the service. In DDoS attacks, the target server faces many demands, not from a specific source, but from different locations of the attack, which makes detection and defense more difficult. With the introduction of network functions virtualization and Software-defined networking, a new route, for network design and management, has been created. The purpose of this research is to investigate and compare DDoS attack defense methods using NFV and SDN. The details provided will help researchers in this field familiarize themselves with DDoS attack defense methods and choose the appropriate design for their actual implementations.

Moral rhetoric plays an important role in our daily situated social decision makings. This importance is more evident during happening of natural and social crises where making the right and fast decisions are urgent. In this research, with the aid of big data generated through Social media which provides a unique opportunity to study behavioral traits and latent psychological constructs, we analyzed the moral beliefs of people during the onset of COVID-19 in Iran. We operationalized Moral Foundation Theory as our theoretical framework for morality and analyzed tweets during the first three months of Covid-19 spreading in Iran. We extracted moral sentiment from tweets using BERT embedding vectors of tweets. We also studied the dynamics of moral belief changes during this crisis. Our analysis showed important features of social discourse among Iranian people during covid-19 which can results to obtain a clearer picture of governing values in discourses about this topic. This picture can help to make timely right decisions to better manage the crises.

Personality tests are one of the tools that can help people to understand themselves better and propose their capabilities and weaknesses. Accordingly, they can choose a suitable career or improve themselves if they are self-aware of their behavioral characteristics. These personality tests speak about what people and their preferences are facing options because games seem to be a solution for studying people’, s personalities by facing them in real-world scenarios and giving them many options to choose from in the gameplay. This research not only investigated the capability of puzzle games as a replacement for classical self-reporting personality tests but also found the correlation of game elements with different personality aspects which are statistically significant compared to personality traits.

Serverless computing is one of the concepts that has become a trend these days. Serverless computing facilitates the deployment of software on the cloud, fog, edge, and IoT infrastructures. Serverless computing helps developers implement their codebase without worrying about server management hassles. These are the reasons why the concept of serverless computing is presented in the cloud, the edge, the fog, and the Internet of things. Research on serverless computing, on the other hand, requires extensive hardware equipments. It can also be time consuming. Simulation is needed to overcome these problems. MFS is a simulator based on the Apache Open Whisk architecture that seeks to facilitate the evaluation process of serverless computing researches. Compared to similar researches, this simulator supports more features including containers, heterogeneous processing resources (e. g., CPU, GPU, and TPU), different runtimes, and Apache OpenWhisk architecture. It also calculates and reports metrics such as average response time, utilization of different computing resources, rejection ratio, missed ratio, number of containers used, and costs. The mentioned metrics can be calculated in overall and individually for each physical machine. The status of all functions is also reported at the end of the simulation (e. g., completed, missed, and rejected).

Improving the performance of recommender systems in the field of product purchase has been one of the important challenges of e-commerce in recent years. The correct recognition of the user's preference and the accurate recommendation of items that meet his needs will certainly lead to progress in the development of online shopping and selling of items. Since the meaning of each item is different for users and the purpose of buying that item changes over time, it is difficult to understand what the user's thinking is now and what kind of product they are looking for. In this work, an approach based on hypergraph neural networks is presented to create a powerful representation for items and users, which can be used to create a correct representation for the user's intent and finally, according to his dynamic intent, offer him/her the right product. The idea of this approach is to use the short-term correlation of items with each other to give them meaning, which is very similar to the logic of buying items by people in the real world. To evaluate the approach, the results of the implementation of the model presented in this article and similar models in recent years are compared on a dataset called Amazon review, and the improvement of the results of the evaluation metrics compared to those models is observed.

Opinions play an essential role in human life. With the ease of sharing opinions, ideas, and feelings on various topics through the web and social networks, the analysis of opinions and emotions has become increasingly important. As social networks continue to expand, the importance of sentiment analysis will only grow. While much research has been conducted on sentiment analysis in the Persian language, its accuracy still falls short compared to available English methods, and it faces several challenges. One of the most significant challenges is the lack of labeled datasets. To improve sentiment analysis, numerous datasets have been collected during various research projects. Despite these efforts, the volume of labeled data remains insignificant because labeling unlabeled data is a costly and time-consuming process due to its manual and human nature. This research presents a semi-automatic method for generating labeled datasets. The proposed method combines pre-trained deep learning models with a human agent, allowing more labeled data to be obtained while spending less money, time, and manpower and using the power of deep learning models. Some unlabeled data were labeled based on this method and added to the basic dataset to create a new dataset called the “, proposed dataset”, . To evaluate the effectiveness of the proposed method, both the basic and proposed datasets were tested on the ParsBERT language model using the same test dataset. The results showed a 4% improvement in ParsBERT 's F1 score on the proposed dataset compared to the basic dataset. Notably, fine-tuning ParsBERT with the new dataset also made it more general and removed one of its weaknesses, i. e., overfitting.

In recent decades, natural language text generation methods have improved for automatic text generation. Also, the recognition and automatic generation of text for personalized persuasion has recently received much attention. The text used to persuade the target audience has been shown to be very effective in various fields. In this paper, we propose to use the CATGAN framework to generate a text that is classified into levels of persuasion and to target people according to the extent to which that persuasion strategy affects each personality trait. We present a persuasive personalized text generation system that achieves state-of-the-art results in text generation for personal persuasion.

In this paper, due to the importance of analyzing users' sentiments in social networks, a hybrid method is proposed in 5 steps. In this regard, after pre-processing the text, the VADER dictionary is used to determine the polarity, then BoW and TF-IDF are used for feature extraction. Also, an ensemble classifier based on logistic regression, decision tree, and Naï, ve Bayes is designed, and finally, several clustering methods are used for aspect-based sentiment analysis. For evaluation, the proposed algorithm is conducted on the Amazon, Twitter, and Reddit datasets the results are reported in terms of run-time and precision.

The recommender system is one of the valuable e-commerce tools. These systems, predict the user's preferences and recommend the products that fit the needs and interests of each user. In classical recommender systems, only the accuracy of recommendations was evaluated. Considering accuracy as the only evaluation criterion leads to problems such as user dissatisfaction due to receiving very similar offers and reducing the probability of purchasing many other products. In new recommender systems, several criteria are used to evaluate the recommenders. One of the evaluation criteria is novelty, which emphasizes that the product is unknown to the user. Converting this definition into a quantitative criterion is challenging. In this research, we proposed a method to create novel recommendations in multi-criteria recommender systems. In this method, the circle of friends is considered as a user's knowledge resources about products. A recommender system is implemented considering the criteria of novelty, accuracy, and coverage and compared with the classical recommender system.

Since its inception in 1990 until today, the Internet has seen many advancements and has evolved over the years. Social networks, video conferencing, 3D virtual world and augmented reality programs are examples of these developments. The term metaverse refers to the next generation of the Internet, a decentralized network of virtual spaces where users can socialize, learn, or play. Technologies play an important role as an enabling enabler for the transition from today's Internet to the metaverse. Technologies such as 5G, blockchain, artificial intelligence, and the transition from two-dimensional graphic space to three-dimensional graphic space, can provide the possibility of creating a interactive and virtual space, equivalent to today's physical world. In this research, the Metaverse architecture, the technologies used and the existing standards in this field have been discussed. Also, the benefits that Metaverse has for people, organizations and governments along with its challenges and risks and the required countermeasures have been examined.

For businesses, discovering key nodes in social networks has a high computational overhead and cost. Because of this, in the problem of influence maximization, at least key nodes with the most influence in social networks are chosen at the lowest possible cost. Numerous algorithms have been proposed to find key nodes in social networks using the technique of influence maximization under the characteristics of submodularity and monotonicity,however, these algorithms struggle with the issue of optimal influence spread because some graph characteristics are ignored when considering diffusion and influence spread in the social network. The HSMD method, which gives special weight to the structure of communities and selects key nodes hierarchically utilizing the structure of communities, has been presented as a solution to this challenge. Based on the characteristics and strategic placement of the nodes within each community, weight is given to each community in this algorithm. Seed nodes are then chosen hierarchically using BFS traversal. This method enhanced the influence spread rate and execution time of recently introduced algorithms such as CTIM, IMBC, RNR, HEDVGreedy, and k-shell.

Descriptive summarization of documents in databases results in better indexing and management of information. Images in documents usually contain valuable information and retrieving them provides tools for document summarization. In context-based image retrieval systems, descriptive tags for the images are extracted from auxiliary information sources available to them. The search engine uses these tags to retrieve images. Here we suggest an automated image tagging method that exclusively relies on information mined from the document’, s text associated with the image. Because of complications in the Persian language, lack of resources, and studies on this language, it has received little attention in the literature. The suggested method is built based on Persian documents. Two groups of tags are suggested. Specific tags are extracted from the caption of the image and the nearby text. General tags are obtained from the keywords of the document. Suggested methods are evaluated on the test data from the Iran scientific information database (GANJ), the largest database of Persian scientific documents. The F-measure of our suggested method is 43% for the specific tags. As for general tags, 89% are descriptive and the false positive rate is 0. 002. Although suggested method has been tested on scientific documents it can be generalized for any type of Persian.

Clustering is the process of partitioning a set of objects into disjoint groups, each partition is called a cluster. Intuitively, it is desirable that the members in each cluster are very similar to each other in terms of their characteristics. As well, it is desirable to have a low degree of similarity between members in different clusters. In general, clustering algorithms can be categorized to follow either a partitioning, a hierarchical, a density, a model-based or any combination of these approaches. The ADBSCAN algorithm is a density-based clustering algorithm which presents a new method to identify high-density local instances considering the properties of the nearest neighbor graph. Two parameters are used in this algorithm, namely the parameter k representing the number of nearest neighbors, and the percentage of noise in the data set. These parameters have a significant effect on the quality of the output as well as the required time. Therefore, it is necessary to find optimal values for these parameters. Brute-force search is one of the naï, ve ways to this end. However, evolutionary-based algorithms such as genetic search methods can be used to make the search process easy and efficient. In this paper, we applied the genetic algorithm to get optimal values of the parameters. The proposed method led to an 11. 46% improvement in the ARI criterion, on average.

The coronavirus, discovered for the first time in December 2019 in Wuhan, China, quickly spread to more than two hundred countries and became a public health emergency. After more than three years, the disease is making a comeback. This article aims to improve the automated detection of COVID-19 infection in CT images by introducing a deep learning-based approach within the context of the Internet of Things (IoT). First, with the assistance of a specialist physician, we collected data in four classes: normal, COVID-19, viral pneumonia, and bacterial pneumonia, and categorized them as COVID and non-COVID. Then, inspired by the image blending, we merged two features of classes, such as COVID-19 and viral pneumonia, to create a new synthetic COVID-19 image. Even though such images occurs infrequently, they plays a crucial role in differential diagnosis. With this idea, the problem of how to collect them will be solved. The DexiNed filter was used to extract deep features from these images, which were then used as inputs to the convolutional neural networks AlexNet, ResNet50, DenseNet201, VGG16, and InceptionV3 for classification. The weighted ensemble of these models achieved an accuracy of 95. 34%, 95% precision, 95. 54% sensitivity, 95. 15% specificity and 95. 27% F1-score.

Network traffic identification is an essential function for network domain systems, which facilitates accurate management through the classification of network traffic flows. In this research we used traffic separation using deep learning approach to detect security anomalies. The method proposed has several steps. Since many features are usually used to detect network security anomalies, in the first stage, feature selection was an optional step to select some of the most important features associated with the problem of detecting security anomalies in the network. Then the SMOTE balancing method was used to balance the data when the evaluated data set was unbalanced in class distribution. The results of balanced data and imbalanced data were obtained. Ultimately, the convolutional neural network was used to train the proposed model. The proposed model was tested and evaluated after training the model. The evaluation results indicated that in the mode of feature reduction and data balancing, the proposed CNN classifier showed the accuracy of 96. 88% and 98. 18% in feature reduction and data imbalance mode, when using no feature reduction and data balancing mode we reached the accuracy of 97. 35% and 98. 57% accuracy in feature reduction and non-balancing data.

This study focuses on the generation of Persian named entity datasets through the application of machine translation on English datasets. The generated datasets were evaluated by experimenting with one monolingual and one multilingual transformer model. Notably, the CoNLL 2003 dataset has achieved the highest F1 score of 85. 11%. In contrast, the WNUT 2017 dataset yielded the lowest F1 score of 40. 02%. The results of this study highlight the potential of machine translation in creating high-quality named entity recognition datasets for low-resource languages like Persian. The study compares the performance of these generated datasets with English named entity recognition systems and provides insights into the effectiveness of machine translation for this task. Additionally, this approach could be used to augment data in low-resource language or create noisy data to make named entity systems more robust and improve them.

Recently, social networks have experienced an exponential growth, providing online users with a venue for expressing and sharing their views in a variety of areas. A very popular social media platform for getting user feedback and collecting data is Twitter. Sentiment analysis consists of extracting and analyzing opinions of people and can be used to predict tweet polarities. In this paper, we present a sentiment analysis methodology for Persian political tweets for the first time in order to assist Iranian politicians. We used two datasets of Persian political tweets with three and seven classes which are labeled according their content. This is the first study of this particular subfield of Persian tweets, so we are using a variety of encoding methods, including Bag-of-words, Word Embeddings, as well as neural methods such as Word2Vec, FastText, and ParsBERT Embeddings. We intend to use these techniques to implement Sentiment Analysis in closed domain political tweets using Machine Learning techniques such as Random Forests, Support Vector Machines, and Neural Networks. As a result of our comparisons, we found that CNN+BiLSTM using ParsBERT embeddings had higher robustness than other networks, scoring 0. 89 on Dataset 1 and 0. 71 on Dataset 2.

In today society, social networking websites have drowned a remarkable attention from users ranging from a child to an old aged person all around the world. The community consumes a huge amount of time on online social networks by interacting and exchanging their information with the other people in the globe. As a result, some of the popular websites like Facebook, Twitter, Instagram, and others witnessed an unexpected growth in registered users. Meanwhile, researches exhibits that all registered accounts are not real,there exist a huge number of fake accounts created for a specific purpose. The major purpose of creating fake accounts is to spread spam content, rumor, and other unauthentic messages on the platforms. This leads to a motivation of developing a system that is able to identify and filter fake accounts on the social networks, but it has many challenges. Researchers have proposed several advanced algorithms to recognize fake accounts. In this paper, the development of fake account detection algorithms using various machine learning approach and deep learning algorithms are reviewed, which give an open vision to the future researchers to develop a foundation in this field.

The digital situation and social networks have introduced a new communication system and created different value hierarchies compared to the pre-digital era. One of the most important new forms of communication system values in the digital situation is the possibility of expressing ideas and feelings and publishing them with the help of social networks. Different modes of expression are provided by digital technology. One of the most historical ways of expression that has changed in its functional forms in the digital situation is the position of writing. The writing position is one of the main modes of expression in the digital situation. Therefore, when the position of writing for a professor of social sciences in a series of occupational and administrative values in the university, in the form of academic writing, is a job necessity, then the position of writing from society for society in social networks for the professor of social sciences becomes a position of playing a role of the popular intellectual. Therefore, the main aim of the article is to form concepts that explain the new position of writing in social networks by a social science professor who is a member of the university faculty in the link between writing as a necessity of a job and an administrative position to writing as a role of a popular intellectual. Based on the conclusion, in the digital situation where it is possible to write and publish at the same time through social networks, the professor of social sciences takes the position of writing-publishing. The position of writing-publishing is related to the post-print era, which provides social science professors with the will to write and publish, and the possibility of writing about society for a large and diverse audience is created. By playing the role of the popular intellectual, the professor of social sciences provides the possibility of propagating the sociological imagination to create awareness of the state of society and this earns the credit of contemporary living for the professor of social sciences.

In recent years, with the growth of science and technology and the creation of competitive markets in various industries measuring the quantitative and qualitative parameters of the final products has become very important. Having a qualified product is the most important part of a production line. Nowadays, there are a few advanced factories that their production is not controlled by intelligent machine vision and image processing-based techniques usually on the Web. Real-time quality management allows increasing production efficiency. The main purpose of this article is to precisely analyze and evaluate some important and influential image processing-based, machine learning-based, and deep learning-based techniques used on the Web that have been employed in the field of quality control. The paper also aims to provide a proper road map for finding the most effective and efficient techniques for the future Web-based projects in the area of quality control.

Due to the growing importance of the cryptocurrency market, as well as the diversity and expansion of online trading platforms, cryptocurrency technology has piqued the curiosity of a wide range of people, from market traders to researchers and analysts. Reliable price prediction is a necessity since investors face multiple challenges including market volatility, risk management, and market complexity. Therefore, numerous studies have been done using deep learning and machine learning algorithms to demonstrate their functionality and efficiency in this area. In this paper, we employed Bitcoin historical data to make predictions for the next day's closing price using a new hybrid 2D-CNNLSTM model with OPTUNA hyperparameter tuning. The dataset used to train the model was gathered using an automated web scraping technique. With the proposed model, the R2 error achieved 0. 98166 and the MAPE was 0. 034. Our proposed model is compared with three different models: CNN, LSTM, and GRU. The predicted results show that the proposed hybrid model is efficient for accurately predicting bitcoin prices and reliable for supporting investors to make their informed investment decisions. Additionally, the proposed model has outperformed other commonly used algorithms, namely CNN, LSTM, and GRU in terms of R2, and MAPE. This model is also capable of performing real-time forecasting.

The convergence of the two mega-trends of "servitization", meaning adding complementary services to products, and "digitalization", meaning the process of using information technology, leads to the provision of services in a new way, known as smart services. It provides the possibility of achieving promising business models. Manufacturing companies that try to provide intelligent services in addition to production, need to plan carefully for their future value network and define their place in it. Service companies should also expand their value creation with new mechanisms in sync with modern technologies such as machine data processing and structures such as digital platforms. Therefore, the complexity of the value network of a successful smart service business is much higher than that of a traditional manufacturing value network. Hence, several studies have been conducted and frameworks have been presented for a better understanding of the value network of businesses, value creation of smart services, and digital servitization. In this article, we review these studies and in this way we investigate the framework consisting of 44 value-creating roles. In the following, to understand how artificial intelligence services affect value perception, consumer participation, and the performance criteria of companies, we will investigate the framework of digital servitization.

Using a mix of machine learning algorithms and big data tools, particularly Apache Spark and also Apache Kafka, this research provides a new method for real-time blood pressure prediction. The method can handle large amounts of inbound data from numerous sources, including wearable technology and internet of things monitors. A clustering-based approach is used to improve the blood pressure estimation's precision while the data is being analyzed in real-time. ECG, PPG, and ABP signals dataset are used to assess the suggested strategy, and the findings show a substantial improvement in blood pressure prediction accuracy when compared to previous methods. The suggested method has the potential to be used in numerous uses, such as remote patient tracking, individualized healthcare, and cardiovascular disease early detection. This research offers two contributions. First off, it introduces a novel technique for real-time blood pressure forecast that is more accurate than current approaches. In addition, it shows the value of merging machine learning techniques with real-time streaming data processing systems like Apache Spark and Apache Kafka. Further improving the scalability and accuracy of the system is the use of web-based tools and deep learning methods. The suggested method may have a big impact on how well patients do and how much it will cost to treat them. Overall, this research offers a path that can be useful to both individuals and healthcare professionals for the creation of real-time blood pressure forecast tools.

In this article, we introduce solutions for solving crossword puzzles by machine using natural language processing techniques. This task is divided into two subtasks of finding possible answers for each table description and then selecting the target word and placing it in the table. The first subtask, which is dedicated to finding the word from its description, has many other uses as in text generation and paraphrasing. For this purpose, we used a combination of different methods, including searching and finding semantic similarities on the data of previously solved tables, searching in dictionary and Wikipedia articles, using a masked language model, and finding related words in Farsnet and the Farsiyar tool. The results show that the combination of these methods has a better result (82% recall) compared to their individual implementation. In the next subtask, we give the list of possible answers to a constraint-satisfaction search algorithm to choose the correct answer that can be placed in the table, taking into account the constraints of the table, and fill the empty cells in the best way and solve the crossword. The overall evaluation shows 80. 22% precision and 68. 86% recall in solving the crossword puzzle.

Spatiotemporal signal processing is one of the complex and hot topics, especially in web mining like web traffic analysis. The web pages and their links are a graph, and their content (e. g., visits) can be a signal. The PyTorch Geometric Temporal is introduced for spatiotemporal signal mining. This study analyzes Wikipedia mathematics pages using the PyTorch Geometric Temporal library to improve their visit prediction during the time using a grid search for hyper-parameter adjustment and analyzing the effect of each parameter. The results show more than 8. 03% relative improvement for the GConvGRU algorithm versus basic related work in state-of-the-art based on about 129, 000 experiments. Besides, it should be considered that lags and node feature parameters must be the same, and lower learning rate and epochs, and higher training ratio and filter size are the best possible values.

To convey complex concepts through speech, learning intelligible pronunciation is essential, which has become more attainable and widespread with the help of modern approaches derived from computer science. Although natural language processing has been widely used in computer science and human-computer communication, its applications in the field of learning and teaching the pronunciation of foreign language words have been largely ignored. In this study, we examine to what extent the implementation of e-learning approaches coupled with automatic speech recognition using trained artificial intelligence helps to understand the progress of language learners' pronunciation skills. For this purpose, 93 English words were selected for pronunciation training in 18 sessions. A common list of 30 words was examined in two experimental groups in two stages of pre-test and post-test to evaluate the improvement of pronunciation of language learners. The information obtained by the language learners was recorded and saved in the form of audio files before and after the teaching process. It was then assessed by automatic speech recognition technology. The results conveyed through automatic speech recognition showed the improvement of word pronunciation skills in the training course.

Aspect-based sentiment analysis (ABSA) is a type of sentiment analysis that aims to identify the polarity of sentiment for aspects in a sentence. Also according to the studies, it is an important research area that plays an important role in business intelligence, marketing and psychology. To solve this problem different methods based on dictionary, machine learning and deep learning have been used. Research shows that among the methods based on deep learning, Transformers has been able to achieve good results and help to understand the language better. In this paper we use induced trees from Fine-tuning pre-trained models (FT-PTMs). We also use dual contrastive learning and different pre-trained models such as BERT, RoBERTa and XLNet in our proposed model. The results obtained from the implementation of the model in SemEval2014 benchmarks confirm the performance of our model.

Influence maximization techniques emphasize selecting a set of influential nodes in order to maximize influence. Because the algorithms presented in this field ignore the topology of cliques for diffusion, there are two major challenges in influence maximization algorithms: optimal diffusion and computational overhead reduction. As a result, the CDP algorithm is presented in this article to address these issues. This algorithm first selects suitable cliques for diffusion based on their position and strategy in social networks. Furthermore, for each clique, a score is calculated based on the topological criteria of the clique, and suitable cliques are selected for diffusion by applying a threshold limit. The seed nodes are chosen from the cliques in the second step. To avoid the rich club phenomenon, only a few nodes from each clique are chosen as seed candidate nodes. Finally, the seed nodes are chosen based on the node's topology and the strength of the node's level one neighbor. In the experiment section, the CDP algorithm significantly outperforms the best algorithms presented in recent years in terms of influence spread rate and execution time.

Today, due to various inccidents (such as stroke, accidents, congenital spinal defects or other reasons) which exist around every person, the number of patients with disabilities or paralysis is increasing, which costs a lot to take care of this type of people in the family. Improving the patient's physical and mental health and converting them to their previous normal life, from a mental and psychological point of view, will be very crucial and has many positive effects on the family and themselves. Hence, using robotic devices can help them obtain their movement in order to use their arms, head and shoulders, which is a great achievement. The aim of this study is to propose a controllable approach for a supervised Fourier-based automatic learning used in rhythmic signals of upper part of the body. The obtained rhythmic signals are analysed by Fourier series and the accuracy of the proposed Fourier series is proved by comparing them to the graphs.

In this study, we analyzed the relationship between the search behavior of users on COVID-19-related queries and the rates of COVID-19 cases. To do this, we first cleaned and visualized data from the COVID-19 Data Repository by the Center for Systems Science and Engineering (CSSE) at Johns Hopkins University, and also gathered data on the most frequently searched COVID-19-related queries using Google Trends. We observed a similarity in the search rates for these queries, which led us to investigate their correlation with COVID-19 cases. To investigate this correlation, we superimposed the search behavior data on top of the COVID-19 case data. We then conducted two statistical tests to further analyze our dataset, which helped us gain insights into the relationship between search behavior and COVID-19 cases. We found a significant relationship between the two variables, which has implications for understanding the public's awareness and response to the COVID-19 pandemic. Our findings suggest that monitoring search behavior may be a useful tool in tracking the spread of the disease and informing public health policies.

One of the most accessible ways to communicate via text is through a short message service. In recent years, profit-seeking people have taken advantage of the good features of this service to send large numbers of spam messages to random people for malicious purposes. In this respect, detecting spam messages is an important task. The unbalanced proportion of the spam and ham data and the extraction of efficient features from short messages have been the main challenges in the SMS spam detection problem. So far, various methods have been proposed to filter spam messages, whose accuracy still needs to be improved. In this study, we propose an ensemble learning method based on random forest and logistic regression algorithms to increase the accuracy of SMS spam detection. The proposed approach has been tested on two real datasets. The experimental evaluation based on accuracy and AUC shows the effectiveness of the proposed ensemble learning algorithm.

In recent years, with the development of science and technology, the emerging competitive markets in various industries, and the need for quality control, quantitative and qualitative measurements of products' properties have become significantly important. Quality production is the most vital part of a production line in a way that a few factories exist, with some of their sectors not being controlled by intelligent Computer Vision (CV) and image processing systems. Accordingly, Real-time quality management can increase production efficiency. Moreover, quality management focuses on customer experience with agents and products. In this regard, Sentiment Analysis (SA) can also determine the success rate of services and products, identify customers' thoughts, and mirror their true voice. This paper aims to conduct a comparative analysis of feature extraction approaches used by CV and Natural Language Processing (NLP) for quality control of products and SA in the industry. These approaches and trends can further be integrated into Web-based quality control systems in industry. Therefore, this paper also investigates how such technologies can enhance quality control through large-scale textual and image data analytics.

In this article, a multi-stage E-commerce platform based on deep learning methods is presented to improve Ford Motor Company's consumer engagement on the Instagram social network. In the first stage, Instagram accounts are crawled and consumers are identified based on comments, hashtags, and followers. With the help of a network built on Darknet as its backbone, the users’,vehicles are detected in the next stage from any shared content. It is in the third stage where MobileNetV2, VGG19, DenseNet121, ShuffleNetV2, and InceptionV3 networks are used to recognize the model and year of each vehicle, and then a voting process takes place to determine the final prediction. In the last step, by applying the provided landmark detection module, all components such as headlights and taillights are localized. In order to improve the robustness and performance of the proposed approach, a heuristic-nested feature extraction block has been embedded at the beginning of each stage and all networks are trained on a dedicated dataset of 16K images in 8 different classes. Based on the results, the vehicles were detected with a mean average precision (mAP) of 72%, and the recognition of each model and year was performed with the accuracy of 98. 14%, precision of 98. 13%, recall of 98. 14%, specificity of 99. 73%, F1-score of 98. 13%, and MCC of 0. 98. The least and the most value of normalized errors (NE) was obtained 0. 0335 and 0. 0504, which corresponds to the hood and right side mirror in the landmark module, respectively.

Nowadays, criminal frauds occur in an organized manner in the banking sector. This issue is challenging since the number of organized frauds associated with such areas is estimated to range from 2% to 5% of the global gross domestic product (GDP). The people committing organized fraud use Internet-based financial services and conventional financial services. Accordingly, they use more complex plans and maps to avoid being recognized through organized fraud fighting systems. Due to the complexity and variety of fraud methods, the transaction may not seem suspicious initially. Hence, it is crucial to consider the interactions between the cards. For this purpose, the use of network theory is recommended. The current paper aims to classify each transaction as illegal or legal correctly. Therefore, extensive data analysis is used to organized fraud in the bank transaction network. Besides, a comparison between supervised learning algorithms is presented on a dataset with 46, 316 transactions related to customers' card activities to distinguish between illegal and legal transactions. According to the Accuracy, Precision, Recall, and F1-Score criteria values, random forest and XGBoost could be considered suitable predictive models for fraud detection.

The purpose of this study is to identify and categorize the legal issues and challenges of information in cyberspace. The current research is a descriptive applied study that was conducted with a systematic review approach using the PRISMA model, content analysis method, and coding techniques. The research community includes 64 English studies and 32 Persian articles and theses. A total of 169 codes were categorized into 17 subcategories, 6 main categories, and 3 themes including interpersonal information management, strategic information management, and effective information management. The theme of interpersonal information management includes the main category of data and information production,The theme of strategic information management includes the main categories of data and information storage and processing and data and information protection,And the theme of effective information management included the main categories of data and information retrieval, data and information dissemination, and data and information use. Sub-categories related to data and information generation including challenges related to ownership, content, tools, technologies, and characteristics of web technology,Subcategories related to data and information storage and processing include challenges related to human factors and process and technology,Subcategories related to data and information protection include challenges related to privacy, data protection, data and information security, protection and security mechanisms, and cybercrimes,Sub-categories related to data and information retrieval include challenges in communicating with users and linking,Sub-categories related to data and information dissemination include challenges related to web technology characteristics and data and information dissemination mechanisms,And the sub-categories related to the use of data and information included challenges related to how to use and limitations of use.

Nowadays, much attention has been devoted to the issues of social networks and social influence. Social influence examines the user's behavioral changes under the influence of their neighbors. The issue of influence maximization is to find a subset of influential nodes that can maximize propagation in the network. The selection of people is very important and is the major aim of the studies. Hence, the current study aims to investigate the maximization of influence in signed social networks since in the psychology of society, negative opinions are superior to positive ones. The criteria considered for measuring influence and methods to increase it by identifying influential people are examined. The proposed solution of this paper is based on the label propagation algorithm. The algorithms used for maximizing influence in signed social networks namely a greedy algorithm and an innovative algorithm are outlined in the second section. To implement the algorithms and simulate the transfer of users' opinions in the graph network, the independent cascade propagation model is used. The proposed algorithm shows better performance and results compared to other algorithms and has less computational overhead since it finds primary nodes by detecting dense parts and not randomly. The significant novelty of the paper lies in the heart of the accuracy and authenticity of the proposed model in maximizing influence in signed social networks.

Getting around CAPTCHAs is essential for stopping fraudulent online activity. The creation of efficient CAPTCHA-breaking algorithms in the context of Persian can help safeguard Farsi-speaking users from a variety of online dangers and enhance their overall online experience. This study offers a novel method for recognizing Persian CAPTCHAs, which was developed and tested on a large and distinctive dataset. Our approach to Farsi CAPTCHA recognition leverages deep learning models, specifically a combination of the TPS-Resnet-BiLSTM-ATTN model, which surpasses other approaches and breaks Farsi CAPTCHAs with the highest possible accuracy. We have achieved amazing results with promising implications for boosting the security and usability of many online services that depend on CAPTCHA authentication by delving deeply into the impact of attention modules on CAPTCHA recognition.

Today, leveraging analytical CRM to maximize values for both customers and businesses is one the most important critical success factors Predicting customer loyalty enables businesses to differentiate among customers for conducting relationship marketing and implementing effective customer extension tactics. In this paper, we analyze customers' reviews on the Digikala e-marketplace to predict their loyalty. We employ NLP, deep learning, and conventional machine learning methods and evaluate the results to find the best prediction model. Two experiments are conducted to evaluate the results: binary and 3-class loyalty prediction. In the binary setting, the Random Forest and Naï, ve Bayes algorithms outperformed the other tested classification methods and achieved an accuracy of 89%. In the 3-class setting, the Random Forest classification method achieved the best performance among all other machine learning algorithms with an accuracy of 67%. The evaluation results imply that businesses could benefit from using the Random Forest classification algorithm to predict customer loyalty through review analysis successfully.

New drug discovery is one of the most important and expensive activities in biomedical research, which is considered as one of the most basic activities in pharmaceutical science. In this regard, pharmaceutical informatics has focused on the drug development with higher quality and lower cost by combining artificial intelligence techniques and basic research of biological and chemical sciences. Considering the importance and capabilities of artificial intelligence in drug discovery, this study is presented with the aim of conducting a bibliometric analysis to identify related studies in this field. Hence, 1030 articles indexed in the WoS database including studies from 1991 to 2023, have been examined. The results show that the topics of artificial intelligence, machine learning, and deep learning were the most repeated keywords in the reviewed articles. In addition, United States of America, China, and India are the countries with the highest number of articles published in this research field, respectively. It was observed that researchers pay more attention to the fields of pharmacology pharmacy and Chemistry multidisciplinary. Also, results demonstrate drug development studies have often been focused on diseases of cancer and diabetic.

Deep neural networks have demonstrated outstanding performance in various image classification tasks, yet action recognition from still images is challenging due to lack of temporal information, scarcity of data, large intra-class variations, and high similarity between different actions. To overcome these problems, ensemble learning appears to be one of the most straightforward approaches. However, ensemble models are computationally inefficient, especially when they are based on a combination of deep neural networks. In order to deal with this issue, we first construct an ensemble model by combining three separately trained deep convolutional neural networks. Then by utilizing this ensemble model as a teacher we train a light-weight student network based on the new knowledge distillation framework. This light-weight model achieves 94. 32% MAP on the Stanford40 dataset, demonstrating its superiority over many existing methods despite its high computational efficiency.

As a country's financing arm, the capital market can play a crucial role in long-term projects and be responsible for large-scale financing. While the development of rail transport has exceptionally positive and significant effects on the country's macroeconomy, and investment in this sector leads to the prosperity of other economic sectors, the return on investment may be delayed or unclear. Consequently, this uncertainty diminishes the investment appeal in this field. Thus, predicting the price and future value of railway company stocks on the stock market is vital due to the need for the foresight to take safe investment measures and achieve sustainable financing. To this end, this paper investigated the use of deep learning methods to forecast the closing price of MAPNA and Toucaril stocks on the Tehran Stock Exchange. Stock prices were predicted using deep neural networks, including Long Short-Term Memory (LSTM), One-dimensional Convolutional Neural Networks (1D-CNN), and CNN-LSTM networks as a hybrid model. Finally, to evaluate the performance of the models, MAE, MSE, RMSE, MAPE, and R2 were used as evaluation criteria. The results indicate that deep learning models can forecast stock prices with accuracy. In this study, CNN-LSTM neural network produced the best results.

This paper deals with the preparation of a universal phone decoder and using it to build a speech-based Persian wake word detection system. First, two sets of spoken data in Persian are used to adjust the symbols and fine-tune the parameters of the phone decoder, which is named Allosaurus, until a wake word detection system with high accuracy is obtained. During this process, a slightly modified version of the Levenshtein Distance algorithm is used to calculate a confidence score for the system output decision. After the initial wake word detector is ready, the values used for calculating the Levenshtein Distance and their weights are optimized in order to achieve the highest possible accuracy. In the end, this work focuses on also maximizing the accuracy of noisy speech signal inputs, which is something that hasn’, t been done in previous works.

Business intelligence (BI) has been considered as a critical factor to provide effective decision making in the organization. But despite the plenty benefits that BI brings to the organizations, it has a limitation in order to analyze qualitative, semi-structured and unstructured data that offers information about customers, competitors, and suppliers. Social Business Intelligence (SBI) is a new concept consists of tools, technologies, applications, and processes that help organizations by collection, processing, and analysis of data types with different structures and formats of different distribution channels such as social media. It creates the possibility of comprehensive monitoring of the environment for the organization. However, a few studies have been conducted on the effective use of SBI in the organization. This research seeks to identify and prioritize the effective factors on the utilization of social business intelligence in the organization. Research data was collected by questionnaire. The research population was academic and business experts familiar with SBI. The snowball method was used for sampling. In the first stage, the factors were extracted from a systematic literature review and validated based on 24 academic and business experts’,comments. Finally, 47 factors were identified and classified into five categories: organizational, media, process, technical and human. In the next step weight of the factors in each dimension was analyzed and prioritized using the Shannon Entropy.

Expanding the use of transformative technologies, artificial intelligence and benefiting from the capabilities of data-driven platforms, has led to the development of concepts, functions and systems of human resources and a comprehensive redefinition of them under the title of digital human resources management. The purpose of this research is to investigate the use of digital in human resource management. After explaining the difference between electronic and digital functions, the researcher has tried to identify the effective components of digital human resources management systems that lead to organizational intelligence with a practical and developmental approach. For this reason, in the first phase 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 an emphasis on the snowball sampling method from the panel of experts. The opinions collected from 14 human resource managers or digital transformation project managers in the studied companies were examined by thematic analysis method and using Max Kyoda software to determine the factors affecting the subject, and in the second phase, the fuzzy Delphi method was used to confirm and finalize the extracted components was executed.

Todays’,data over web are retrieved from Relational Database Management Systems (RDBMS). Web retrieval and content analysis by keyword search in relational database has become the preferential way to search in todays’,era of Big Data. Any keyword search system has two main parts including candidate network generation and evaluation. Most research’, s attention is on providing an efficient candidate network evaluation algorithm and relies on non-efficient candidate network generation paradigm. In this work, a novel probabilistic candidate network generation algorithm is proposed. From the outcome of our investigation on benchmark databases including new ISC database, it is possible to conclude that the proposed candidate network generation approach enhances the efficiency of databases keyword search.

Due to drastically increasing data and increasing productivity which can be obtained, processing and analyzing big data plays a crucial role nowadays. On the other hand, a growing range of standardization organizations has addressed various challenges in the field of big data, which are important in design and development of big data systems and services. The big data reference architecture is developed by different standardization institutes including joint technical committee of the International Organization for Standardization (ISO/IEC), the American National Institute of Standards and Technology (NIST), and also the International Telecommunication Union (ITU). The reference architecture describes the structure and components of big data from different perspectives, including conceptual, user, and functional. This paper addresses the most important related standards and provides a mapping of them to the big data reference architecture, which could be used for various purposes, mainly as a standardization roadmap. This road map could facilitate various activities, including technology-level assessment, testing and comparing capabilities, type approval, and licensing.

Traceability and auditability are essential features in supply chain management and construction. However, from the customers' perspective, trust is the most critical aspect of these systems. Additionally, relying on third parties for trade in centralized systems is indispensable. Blockchain-based drug traceability provides a potential solution to create a platform for an immutable, trustworthy, accountable, and transparent system in the pharmaceutical supply chain. Furthermore, we present a model for storing pharmaceutical supply chain data using blockchain, which leverages the key advantages of blockchain and smart contracts. Our proposed solution is based on Hyperledger Besu and is complete and cost-effective in terms of the privacy and confidentiality of the blockchain network. Additionally, our proposed model addresses the challenges of storing large data in the blockchain, which we intend to solve by using the IPFS storage system. In such a system, instead of storing big data, we store their abstracts in the blockchain to reduce the load pressure of the chain and realize efficient information queries.

Multiple sclerosis (MS) is an autoimmune disease that causes physical disabilities over time. The prevalence of this disease is increasing in the world. Since this disease causes physical disabilities, the prognosis of relapse is vital. Currently, doctors use magnetic resonance images to diagnose the disease and its process. But this method suffers from some problems such as operator fatigue, the difficulty of using magnetic resonance images, the complications of its use, etc. For this reason, machine learning techniques can be used as diagnostic aids for doctors. Some data were from the hospital’, s HIS and electronic health records. Some other physical files of patients have been used. Overall, 492 records of 84 patients with relapsing-remitting type were collected. We studied the clinical data of patients with disease onset for two years. The data were collected from the MS clinic of Imam Hossein Hospital in Tehran. The patient's clinical symptoms were recorded from February 2021 to February 2023. Mainly, the patient's symptoms were classified into 7 groups of relapses. Also, the patient’, s personal information and relapse symptoms were examined. With the help of machine learning techniques, we will design a model that predicts the probability of relapse in the next year. The data were unbalanced, so the SMOTE oversampling method has been used. The results show that these models predicted relapse in patients with high accuracy and precision. This research used standard machine learning models SVC, K-NN, Logistic Regression, Decision Tree, Random Forest, Naive Bayes, MLP, XGBoost, Ada Boost, and Gradient Boost. The results of evaluating the models based on accuracy and precision criteria showed that the SVC with 98. 837% precision and 97. 973% accuracy and followed by 97. 674% precision and 97. 297% accuracy in both LogisticRegression and MLP models are the best model and can provide an appropriate prognosis.

Machine translation is a technology that reduces costs and speeds up the translation for users by mechanizing translation from one language to another. Machine translation is essentially a step towards globalization for every culture, science, industry, and system. This technology has made great strides in cognitive understanding of natural language since 2013 with the advent of deep learning models. But deep models also always need a lot of data for training. Therefore, the lack of a lot of data and parallel corpus in machine translation is one of the most important problems in this area. Machine translation from English to Persian always suffers from the problem of a lack of resources and data. This article tries to study the deep learning models in machine translation from English to Persian and their strengths and weaknesses, In this article, to solve the problem of lack of English-to-Persian data, the Transformer-based model has been integrated and improved with the Persian language model that has GPT architecture, in addition, the CNN model has been integrated and improved with Autoencoder to improve feature selection and reduce dimensions.

The crowdsourcing platform is a very popular platform where individuals collaborate to solve complex problems. Requesters share tasks that they are unable to solve, and workers perform these tasks based on their expertise. In today's digital world, privacy is a major concern for users of this platform. Most workers prefer to keep their personal information, such as age, location, and interests, private. Additionally, ensuring that workers receive rewards after completing their tasks is another reason for participating in crowdsourcing. However, traditional crowdsourcing requires a third party to send the rewards to workers, which can be unreliable and destructive. Therefore, most workers prefer platforms that guarantee protection of their privacy and full payment of their rewards. In this article, we use blockchain technology and smart contracts based on Ethereum to eliminate the need for an unreliable third party and make workers fully assured that they will receive their full rewards. In addition, we improve worker privacy protection using encryption algorithms and show that worker privacy can be largely protected by a feature-based encryption algorithm with a ciphertext policy.

Currently, a lot of worry and anxiety among people both locally and internationally about the coronavirus potentially becoming a pandemic and spreading globally. However, in recent years, more and more people have been relying on social networks as their primary source of news and information. As a result, many political figures are greatly concerned about the widespread false and deceptive information on social media. We are not only facing the challenge of combating COVID-19 but also an "infodemic. " To tackle this issue, A dataset consisting of 7, 000 social media posts in Persian related to COVID-19 has been collected and made available. This data consists of both true and false news. Several languages, such as Arabic, English, Chinese, and Hindi, have also recognized the problem of fake news related to COVID-19. In this study, a deep neural network approach is used to simplify feature extraction, develop a strong ability to learn, and automatically discover features, which is more effective than traditional machine learning approaches. Additionally, a novel approach to improving outcomes using a deep neural network is employed. The hyper-parameters of the deep learning algorithm are set and optimized using genetic algorithms and Q-learning, resulting in better outcomes than previous research and achieving an accuracy rate of 0. 92 percent.

Developing indoor positioning systems (IPS) using Received Signal Strength (RSS) received from Wi-Fi sensors based on the Fingerprinting technique has become an interesting topic to many researchers in the positioning and navigation field. These systems, however, are barely tested in real environments such as complex buildings and in real situations where the change of RSS over time occurs frequently. Moreover, the cases tested in an environment close to a real scenario could not provide proper positioning accuracy. In this paper, some methods are proposed to improve the IPS with the KNN algorithm which changes the procedure of selecting the chosen train points. These methods are, also, implemented and tested in a real building with different complexities and an area of 324 m2. The results show an improvement of 14% in average positioning accuracy from 2. 11 m to 1. 82 m.

In order to improve the policy-making process in any governance system, including the content of cyberspace, it is necessary to examine the function of various institutions of the system. Doing this helps to extract inconsistencies, overlaps and conflicts between the functions of policy-making institutions and their policy programs in the field of cyberspace content and provides a proper understanding of the specific institutional arrangement of a governance system. In this regard, this research seeks to identify the policy functions of policymakers in the cyberspace content governance system. In this research, by reviewing the existing writings in the field of cyberspace content policymaking and using a meta-synthesis qualitative approach, a set of policy functions of the cyberspace content policymaking system are identified. Based on the results of this research, the functions of policymakers in the cyberspace content governance system are: policy-making, legislation, monitoring and evaluation, development and strengthening of education and research, culture making and promotion, support, business environment development, organizing licenses and permissions, and regulatory.

This article is particularly devoted to underlying information and industrial technologies and the conceptual basis for the Industry 4. 0 and Industry 5. 0 concepts. The report shows that the Base Concepts, Underlying Information, and Industrial Technologies for the Industry 4. 0 Concept are the technological platforms, which are the transformations of technologies reached over the previous stages of industrial development. The report also shows that human-computer interaction is a technological bridge leading from Industry 4. 0 to Industry 5. 0. A practical example of the computer control system’, s engineering intended for protection against unauthorized physical access to the data server is also given in this paper. The presented control system implementation includes human-computer interaction, the Internet of Things, and cyber-physical systems and has the properties of Industry 4. 0 and Industry 5. 0 systems.

Multiple sclerosis is an autoimmune disease that causes physical disability. There is currently no definitive treatment for the disease. Immunosuppressive drugs are used to reduce recurrence and delay disability. The advances in information technology have expanded the use of artificial intelligence systems, including recommendation systems. One of the applications of medical recommender systems is prognosis, diagnosis, and treatment. This study used the data of relapsing-remitting multiple sclerosis (RRMS) patients collected from the MS clinic of Imam Hossein Hospital in Tehran. The data of new patients who are women and aged below 40 years and above 18 years were used. We intend to use clustering and the K-Means method in this study. Also, using cosine similarity, we offer recommendations for a cluster that resembles a new patient. The collaborative filter approach is implemented as one of the recommendation system methods. In other words, a pharmaceutical recommendation system is provided for patients with MS. The results of this study show that the average precision is 98. 198%, and the average Recall is 97. 756%. Therefore, it performs well for the recommended system.

Nowadays, most of our daily activities are carried out on the web. The high speed and volume of data production on the web have made the use of online machine learning algorithms in processing and analyzing data streams very efficient. Many of these algorithms have been developed assuming a fixed feature space,however, in real-world problems, this assumption may not hold and each instance of a data stream may have different features. In this study, this new problem that has recently attracted a lot of attention is investigated. Also, a novel general algorithm for data stream classification is proposed, which exploits the relationships between features and estimates the values of unavailable features to achieve the maximum potential classifier. Finally, through empirical experiments and comparison with two recent algorithms, it is shown that the proposed algorithm has higher accuracy.

In Persian, the grammatical particle ezafe connects two words. Ezafe is one of the salient factors in Persian phonology and morphology to understand the meaning of a sentence completely and truly, whereas it is not usually written in sentences, resulting in mistakes in reading complex sentences and errors in natural language processing tasks. Therefore, recognizing words that need Ezafe at the end of themselves, is a major factor to improve the performance of a variety of NLP-based systems such as a Text TTSsystem. Because in Persian TTS systems without an Ezafe recognition module cannot make Ezafe constructions to read the text correctly and does not recognize the relations between the words. As Transformer-based methods shows state-of-the-art results in lots of NLP tasks, in this paper, we experiment ParsBERT in the task of ezafe recognition. The latter earning 2. 68% better F1-score than the prior state-of-the-art, we obtain the most advantageous outcomes.

Over the recent years, the adoption of Mobile Edge Computing (MEC) has increased due to its ability to bring computing resources closer to end-users, which includes storage, computing, and networking at the network's edge. This approach results in faster and more efficient data processing, reduced latency, and better overall performance for mobile device applications. Our aim in this study is to evaluate the effectiveness of using reinforcement learning algorithms, namely Deep Q-Network (DQN) and Asynchronous Advantage Actor-Critic (A3C), in optimizing the performance of web applications in MEC environments, such as latency, CPU usage, and memory utilization. We conducted experiments using a sample dataset and compared the performance of models with and without MEC. The results demonstrate that the use of MEC substantially improves the performance of web applications. Both DQN and A3C algorithms exhibit promising results in improving the latency of web applications in MEC environments. However, the A3C algorithm outperforms the DQN algorithm in terms of CPU utilization and memory usage. Overall, our study highlights the potential of reinforcement learning algorithms in improving the performance of MEC-based web applications.

Cloud data centers are a model of distributed systems that provide users with services to share data and information through the Internet. These centers also face challenges due to their popularity among users. With the increase in the number of hosts to respond to users' needs, challenges such as increasing power consumption, service level agreement violation, and time are seen. As a result, it is very important to address these challenges in these centers in order to reduce costs and increase profits. task scheduling for hosts is one of the most effective methods to improve productivity and optimal use of hosts' resources. In this process, with proper allocation, we can prevent hosts from becoming overloaded and increasing energy consumption due to inefficient use of hosts' resources. The proposed solution in this paper is to use multiple goals for the allocation process using the Crow search optimization algorithm. The Crow search optimization algorithm is new, fast, and powerful. As a result, in the proposed method by modeling this algorithm and considering the multi-criteria fitness function based on the requested resources of the tasks and the available resources of the host, we tend to manage resources properly. The simulation results show that the proposed method has a 9% reduction in service quality parameters such as power consumption compared to paper [21] and 15% compared to article [16], 11% execution time compared to paper [21], and 14% compared to the paper [16] and the service level agreement violation has improved by 16% compared to the paper [21] and 8% compared to the paper [16] and has been able to reduce the mentioned parameters.

Every day the modern world is moving towards digitalization and cashless transactions are becoming more common, credit cards are rapidly becoming more popular. Online and offline purchases using credit cards have become increasingly popular, which results in more fraudulent transactions every day. A large number of credit card fraud incidents occur every year and lead to huge financial losses. accordingly, it could be important to choose the best fraud detection method is essential so that it can detect fraud before criminal consumers a stolen card. To detect fraud, one method is to evaluate historical transaction data, as well as both normal and fraudulent transactions, to obtain usual and fraudulent behavior features by using machine learning techniques. we can use machine learning algorithms to solve this problem if we have access to enough data. In this study, our goal is to compare three algorithms for detecting credit card fraud (Decision Tree, Regression Logistic and Random Forest). we want to use a model that is new and based on a hybrid approach for detecting credit card fraud. According to this study, the proposed model is more capable of identifying fraudulent transactions than previous studies.

The ever-increasing expansion of the volume and variety of data and their role in the creation of strategic insights has made the use of analytical tools and the implementation of data-driven marketing in marketing as a priority for chief marketing officers (CMOs). Despite increasing interest toward data-driven marketing (DDM) approach by practetionares and academics, the implementation of these analytical initiatives does not achieve the promised results. One of the reasons for the failure of analytical efforts is the lack of a roadmap and a DDM strategic planning methodology that assissts CMOs in implementing and aligning such emerging initiatives with business strategies. However, limited studies have addressed strategic topics and especially strategic planning in the field of data-driven marketing. Therefore, the purpose of the research is to design an integrated and novel methodology for strategic planning of data-driven marketing. To achieve the goal of the research, a qualitative approach has been used. Along with the comprehensive litrature review, the focus group was used to explore the dimensions and activities of the proposed methodology. The findings of this study show that the main phases of the data-oriented marketing strategic planning methodology include Determine Strategic position, Strategic contextualization for DDM, Strategy development, Action plan development, Performance management. In the proposed methodology, the main phases and activities related to each of the phases and the sequence of their implementation are presented in a coherent roadmap for applying DDM with a strategic perspective to leverage analytics tools in realizing the long-term promises of analytical initiatives.

Cloud computing has grown exponentially as an essential technology for the development of Internet services these days. The spread of this technology has led to the creation of large-scale data centers and complex practical applications with a wide range of communication needs. There has been an increased demand for cloud services in recent decades which led to higher energy consumption and Service Level Agreement Violation (SLAV). Through utilizing the Service Level Agreement Violation (SLAV) parameter, it is possible to calculate the SLA of a data center. Via reducing the number of migrations of Virtual Machine (VM) aggregation, we can establish the SLA and reduce the SLAV accordingly. In this study, we have proposed a suitable method per consolidation for choosing an appropriate VM to migrate from an overload host to a destination host using fuzzy logic. For doing so, a variety of VM selection policies for migration, such as Minimum Migration Time (MMT), Maximum Correlation (MC), migration control, and Maximum Power Reduction (MPR) have been fuzzified,hence different parameters in selecting a VM for migration have been considered. Using CloudSim simulator was indicated that the proposed approach had made improvements in the case of the number of migrations, total SLAV, SLA time per active host, and energy consumption. Using the migration policies fuzzification, we reduced SLAV and energy consumption.

The tourism industry has undergone a significant shift towards data-driven strategies in recent years. As a means of improving the quality of their service and performance, service providers are analyzing feedback from their customers to increase the number of tourists they attract. Negative feedback also provides valuable insights into the factors that detract from a location's appeal. Datasets that gather information on people's experiences and opinions of tourist destinations can be analyzed to extract valuable information. However, there are currently few existing datasets that specifically capture user reviews about historical and tourist attractions in Iran. To fill this gap, users have shared their travel experiences on various websites, and sentiment analysis can be employed to extract insights from this data. Effective sentiment analysis requires a suitable approach for data extraction, pre-processing, and storage. This study provides a framework for the user review dataset preparation, including data collection, ETL, data storage, and evaluation phases. A rich dataset containing user reviews about 178 Iran's historical and tourist attractions was prepared through the proposed framework in which automated crawlers were developed to collect data from Tripadvisor platforms. Data labelling was achieved using the DistilBERT-base-uncased language model for sentiment analysis and human evaluators for final annotations. A total of approximately 25 thousand samples were included in the dataset, and positive user comments outnumbered negative user comments by a wide margin. This high percentage of positive comments suggests that the locations were of a satisfactory standard, making it likely that users would return in the future. The findings of this study can help providers to improve the overall quality of their services by analyzing user reviews. The proposed framework and achieved dataset can also guide future efforts to leverage data for improved performance and customer satisfaction in the tourism industry by identifying areas that need improvement.

In recent years, billions of users are exchanging information at an increasing rate in the vast and interconnected world of the Internet. On the other hand, attackers keep planning various threats in this environment. As a result of these issues, it has become more and more critical to prevent or reduce vulnerabilities. Cross Site Scripting is a well-known vulnerability on the web through which an intruder tries to steal users’,vital information or induce malicious activities on behalf of the user. The weakness of the previous methods of detecting XSS attacks based on machine learning lies in not taking the possible changes in the characters of the attack vector into account, or in other words, special encodings, and this reduces the accuracy of these methods. The method presented in this article utilizes an algorithm that translates a kind of obfuscation in the attack vector that increases the accuracy of the detection model on the XSSED dataset to over 98%.

The evolution of Tehran subway networks has been investigated from a complex network perspective in order to analyze their structure, topology and functions. Tehran Metro networks are characterized by a large number of interconnected stations, which makes them highly dynamic and nonlinear systems. From a complex network perspective, The stations are the network's nodes, and the lines connecting them are its edges. Metro networks display scale-free and small-world features, according to previous investigations, with preferential attachment forming the basis for their formation. Additionally, the evolution of Tehran metro networks is driven by transportation demand, economic activities, geography and urban planning. This study shows that the better network has been obtained during the evolution as the degree, clustering coefficient, and the shortest path length have been improved in the third phase but network robustness is not so high against to the attacks.

L-Atur is a web application start-up that produces highly customizable generative designs for consumers at affordable prices. It uses L-Systems as its vital core to generate a wide range of designs. L-Systems can provide numerous visual responses to the creative exploration of our users. Consumers would be able to design their desired products without any coding or professional design skills. The human and machine collaborative nature of L-Atur helps them to explore creative opportunities while we consider technical limitations to physically produce their designs. The final results are based on the countless visual possibilities of L-Systems, so we can make one-of-a-kind designs accessible to a wide range of people.

Social networks play a significant role in our lives, serving as a platform for the exchange of views, thoughts, and opinions. Consequently, sentiment analysis has become a valuable process for collecting and analyzing people's opinions on a wide range of issues. Given the global COVID-19 pandemic and the development of various vaccines to combat it, people have expressed a range of opinions on Twitter. By analyzing these opinions, health organizations can become more aware of people's feedback and emotions. Despite the benefits of sentiment analysis, challenges remain with respect to accurately interpreting and determining the appropriate polarity of sentiments. These issues may negatively impact people's thoughts and opinions when it comes to making informed decisions. To address this issue, we developed a model for classifying people's tweets into three categories: positive, negative, and neutral. We used a sizable dataset extracted from Twitter comprising 228, 207 tweets and an architecture based on LSTM and BiLSTM-CNN models. The results obtained from the experiment indicate that each model could achieve 93. 66% and 94. 10% rates, respectively, which outperformed the other models.