مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119A Hybrid Deep Learning Model Based on BLSTM-GRU for Enhanced Call Volume Forecasting in Call CentersA Hybrid Deep Learning Model Based on BLSTM-GRU for Enhanced Call Volume Forecasting in Call Centers182195enArmaghanNikfarDepartment of Computer Engineering, Ka.C., Islamic Azad University, Karaj, Iran SasanHarifiDepartment of Computer Engineering, Ka.C., Islamic Azad University, Karaj, Iranhttps://orcid.org/0000-0002-6788-8222JavadMohammadzadehDepartment of Computer Engineering, Ka.C., Islamic Azad University, Karaj, Iranhttps://orcid.org/0000-0003-1889-0294202625This research aims to develop a forecasting model for call center call volume by incorporating characteristics unique to each country and call center environment. Using historical call data, calendar events, and other relevant input variables, the goal is to create a framework that is capable of accurately simulating and predicting the volume of incoming calls over specified time periods. For this purpose, we employ advanced recurrent neural network architectures, including MLP, RNN, LSTM, and a hybrid BLSTM-GRU model, to analyze temporal patterns in call data. The BLSTM-GRU model, in particular, uses two-way and gating processing mechanisms to more effectively capture complex temporal dependencies. The dataset consists of daily call volume over a 4-year period along with text features formatted in Excel for ease of training and evaluation. Furthermore, the forecasting framework integrates a comparison mechanism that compares the predicted call volume against the actual values to assess the accuracy and robustness of the model. Experimental results show that the BLSTM-GRU model outperforms standard architectures, achieving a forecasting accuracy of 94.12%, whereas the optimized single LSTM baseline reaches 81.35%. To increase operational efficiency, the framework also generates shift schedules that comply with call center standards and prevent call queues by recommending the required number of agents. The superior performance of the proposed BLSTM-GRU model highlights its potential application in various contact centers around the world and provides a scalable solution for intelligent contact center management.

http://jist.ir/fa/Article/Download/52928

مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119Neural Risk-Adaptive Access Control: A Behavior-Context Hybrid Framework for Adaptive Authorization in Dynamic EnvironmentsNeural Risk-Adaptive Access Control: A Behavior-Context Hybrid Framework for Adaptive Authorization in Dynamic Environments196213enDonyaPiriaeiDepartment of Computer Engineering, Bo. C., Islamic Azad University, Borujerd, IranAfshinRezakhaniDepartment of Computer Engineering, Faculty of Engineering, Ayatollah Boroujerdi University, Borujerd, IranHamidHaj Seyyed JavadiDepartment of Computer Engineering, Shahed University, Tehran, Iran0000-0003-0082-036XLeilaRikhtechi Department of Computer Engineering, Bo. C., Islamic Azad University, Borujerd, Iran202668We present a risk-adaptive access control framework that integrates dynamic behavioral modeling with Attribute-Based Access Control (ABAC) to enable real-time decision-making in volatile and high-risk environments. The proposed system introduces two novel components: the Attention-based Temporal Unit (ATU) for capturing sequential access behaviors, and the Contextual Scoring Module (CSM) for quantifying contextual and environmental signals. Risk is estimated via a multi-dimensional function that combines behavioral patterns, device and location attributes, and real-time threat indicators. An adaptive decision threshold, continuously refined based on false-accept and false-reject outcomes, ensures a calibrated balance between security and operational usability. Computed risk scores are seamlessly embedded into ABAC policies to support fine-grained enforcement. Evaluation on four domain-relevant datasets—Login Data Set for Risk-Based Authentication (RBA), CERT Insider Threat Test Dataset, CIC-IDS2017, and RT-IoT2022—used to emulate access control scenarios, shows F1-scores up to 0.95 and precision up to 0.94, outperforming strong baselines by up to 12%. The results demonstrate robustness across authentication, insider threat, intrusion, and IoT domains, supporting the framework’s applicability in real-time, Risk-Adaptive access systems for wireless communication and IoT environments, particularly in emerging 5G/6G contexts. The modular architecture also separates behavioral, contextual, and environmental evidence so that each source can contribute to the final authorization decision without requiring a single static risk signal.

http://jist.ir/fa/Article/Download/53783

مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119Fusion-Based Deep Learning Framework Detection Using an Optimized for Cardiovascular Disease Squeeze Net ModelFusion-Based Deep Learning Framework Detection Using an Optimized for Cardiovascular Disease Squeeze Net Model214224enM Shanmuga SundariDepartment of Computer Science and Engineering, BVRIT HYDERABAD College of Engineering for Women, Hyderabad, India0000-0001-5755-474X202527The electrocardiogram (ECG), which measures the electrical activity of the heart, is frequently used to identify heart conditions because it is simple and non-invasive. The primary focus of the analysis is on the fusion of action impulse patterns produced by the heart’s specialized cardiac tissues. It is possible to detect any irregularities in the heart by closely examining the electrical signal generated with each heartbeat. Since the application of data mining technology in healthcare has greatly increased knowledge, data mining is becoming a more and more popular option in the medical field. This study uses a hybridized algorithm for pattern mining in order to categorize heartbeats. First, the input data from the ECG signal is pre- processed using a median filter, and then the features are extracted. Both statistical and medical aspects are included in these extracted features. Activity pattern mining using the Optimization Search Algorithm (OSA), a hybridized optimization algorithm, is the next step. Lastly, the rule matrix produced by the mining of activity patterns is used to categorize heartbeats. This research proposes a deep learning-based framework for the classification of cardiovascular diseases using electrocardiogram (ECG) signals. A hybrid optimization technique, integrating the Taylor series and Ebola Optimization Search Algorithm (OSA), termed OSA-SN, is introduced to enhance the performance of a lightweight convolutional neural network model, SqueezeNet. The ECG signals undergo preprocessing and feature extraction, followed by sequential pattern mining using the proposed optimization strategy. The optimized model achieves superior performance metrics, with a precision of 0.93, sensitivity of 0.94, and specificity of 0.91. These results demonstrate the potential of the OSA-SN-enhanced SqueezeNet model in accurate, real-time heartbeat classification for clinical applications.

http://jist.ir/fa/Article/Download/49428

مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119Soft and Safe Network Slicing with Hierarchical Multi-Agent Reinforcement Learning225237enZeinabSasanAmirkabir University of TechnologysiavashkhorsandiAmirkabir University of Technology0000-0002-2802-59422025106Network slicing has emerged as a key enabler for next-generation mobile networks, allowing multiple virtual network slices to coexist on shared physical infrastructure. Achieving high utilization while ensuring low dissatisfaction and safe operation across slices remains a challenging task, especially under dynamic and heterogeneous demands. In this paper, we propose a Hierarchical Multi-Agent Reinforcement Learning (HMARL) framework for soft and safe network slicing resource allocation. Our approach leverages a two-level learning architecture in which local agents manage slice-level decisions while a global agent coordinates overall resource distribution. At the lower layer, slice-level agents observe local conditions such as pending requests, deadlines, and resource demands and make admission or scheduling decisions accordingly. A meta-controller at the higher layer, adjusts inter-slice capacities and resource boosts to balance utilization, fairness, and safety across the entire system. We compare HMARL against Flat MARL, No-Sharing, and Random admission baselines to highlight its performance advantages. Extensive experiments show that HMARL achieves higher utilization and better objective function values while significantly reducing rejection ratios and dissatisfaction compared to other approaches. These results demonstrate that hierarchical reinforcement learning can provide a scalable, flexible, and safety-aware solution for resource management in soft network slicing environments improving both scalability and adaptability.

http://jist.ir/fa/Article/Download/51727

مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119Assessing the Usability of a Comprehensive Research Management System: A Mixed-Use Study in Iran238244enMehdiBeyramijamDept. of Health in Emergency and Disaster, School of Public Health, University of Medical Sciences, Hamadan, Iran.0000-0002-1373-7808Mohammad SadeghSohrabiOccupational Health and Safety Research Center, Institute of Health Sciences and Technology, Hamadan University of Medical Sciences, Hamadan, Iran000000023543578XمحمدبابامیریDepartment of Ergonomics, School of Public Health, Research Center for Health Sciences, Institute of Health Sciences and Technology, Hamadan University of Medical Sciences, Hamadan, Iran.0000-0002-0824-8567ArezoFarhadiPh.D. student, Department of Medical Library and Information Science, School of Health Management and Information Sciences, Iran University of Medical Sciences, Tehran, Iran.0000-0002-3458-69062026215Background and Objective: Usability is a key dimension of the quality of research systems. This study aimed to explain usability and examine users' experiences and challenges in interacting with the "Pajouhshiyar" research affairs management system at several medical universities in Iran. Methods: A convergent parallel mixed-methods design was employed. Quantitatively, 169 users (76 men, 93 women) were assessed using a questionnaire for evaluating the usability of web-based software; independent t-tests examined differences by gender and by role. Qualitatively, 16 semi-structured interviews were conducted via purposive sampling and analyzed using content analysis. Results: The mean “Relative Usability” score was 33.19 % (low to moderate). The lowest subscale was “Ease of Use” (23.82 %), and the highest was “Content/Output Quality” (42.40 %). In between-group comparisons, women scored higher than men across most domains (no significant differences for “System Terminology and Information” and “System Capabilities”). Likewise, officers/managers outscored researchers on most scales (no significant differences for “System Terminology and Information,” “Personalization,” and “Error Management”). The qualitative analysis yielded five main categories: complex pathways/menus and similarity of options; system errors and saving/storage problems; insufficient training and support; strengths in documentation and national integration; and improvement suggestions including process simplification, redesign of search/reporting, personalized dashboards, and attention to a mobile version. Conclusions: Despite strengths in documentation and integration, deficits in ease of use and learnability hinder optimal efficiency and satisfaction. User-centered workflow redesign, improved error handling and feedback, structured training, and continuous user experience monitoring could enhance system adoption, user satisfaction, and productivity.

http://jist.ir/fa/Article/Download/53112

مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119A Hybrid Entity-Centric Approach to Persian Pronoun Resolution with BERT Enhancement245267enHassanHaji MohammadiDepartment of Electrical and Computer Engineering, Islamic Azad University Tehran North Branch, Tehran, Iranalizeratalebpourshahid Beheshti0000-0003-2538-3928Ahmad Mahmoudi-AznavehCyberspace research institute, Shahid Beheshti University, Tehran, IranSamanehYazdaniDepartment of Electrical and Computer Engineering, Islamic Azad University Tehran North Branch, Tehran, Iran2025131Pronoun resolution, a critical aspect of coreference resolution in natural language processing, is essential for accurate text understanding and generation. This paper introduces a hybrid Persian pronoun resolution model that combines rule-based and machine-learning approaches to address the unique challenges of Persian linguistics. The model integrates seven high-precision rule-based modules, specifically designed for Persian linguistic nuances, with a random forest classifier enhanced by BERT contextual embeddings and cluster-level features. This hybrid approach effectively tackles the complexities of pronoun resolution, outperforming existing Persian systems on two corpora: the newly developed Mehr corpus (400 documents) and the established RCDAT corpus. The integration of BERT embeddings and cluster-level features plays a pivotal role in the model's superior performance, yielding a 4.68 F1-score improvement on the Mehr corpus and a 4.01 F1-score improvement on the RCDAT corpus. The proposed model not only achieves state-of-the-art results but also provides a practical and interpretable solution for Persian pronoun resolution. By leveraging both rule-based and machine-learning techniques, this work advances the field of Persian natural language processing, offering a robust framework for future research and applications. The findings highlight the importance of combining linguistic rules with advanced machine-learning methods to address the intricacies of pronoun resolution in morphologically rich languages like Persian. This study contributes significantly to the growing body of research on Persian NLP, paving the way for improved text understanding and generation in Persian.

http://jist.ir/fa/Article/Download/49358

مرکز منطقه ای اطلاع رسانی علوم و فناوریJournal of Information Systems and Telecommunication (JIST) 2322-143714552026119Hybrid Embedding and GPT-5 Instruction-Augmented Framework for Arabic Hate Speech Detection268276en Anas Khudhur Abbas al jubooriUniversity of Tehran, Department Pardis Alborz, Computer EngineeringHeshaamFailiTehran University20251028Hate speech refers to expressions that incite violence, hostility, or discrimination against individuals or groups based on attributes such as religion, ethnicity, gender, or national origin. With the rapid growth of social media, hate speech and abusive language have become increasingly widespread, raising concerns about user safety, social cohesion, and responsible online communication. Detecting such content is highly challenging due to the massive scale of user-generated posts, the informal and evolving nature of online dialects, and the limitations of current automated systems. While significant progress has been achieved in high-resource languages such as English, research on Arabic, particularly its dialectal varieties, remains limited despite its global importance. This study addresses these gaps by proposing a comprehensive methodology for hate speech and abusive language detection in Arabic. We adopt hybrid embeddings by combining AraBERT-v2 with Multilingual MPNet to capture both dialect-specific and cross-lingual semantic features. Dimensionality reduction with UMAP and clustering using HDBSCAN are refined iteratively through Silhouette and Coherence scores. From these clusters, linguistic grounded instructions are generated with GPT-5 and evaluated using a two-stage process: LLM-as-a-Judge prompt evaluations for automated assessment, followed by human validation to eliminate ambiguous cases. The refined instructions are then used to fine-tune a classifier on the L-HSAB dataset. The final instruction-augmented model achieved over 92% accuracy, demonstrating that combining hybrid embeddings, adaptive clustering, and instruction-based fine-tuning provides an effective framework for dialectal Arabic hate speech detection and can be extended to other languages.

http://jist.ir/fa/Article/Download/51935