Egyptian Informatics Journal最新文献_第6页

Arabic hate speech detection based on BERT models variants 基于BERT模型变体的阿拉伯语仇恨语音检测

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-20 DOI: 10.1016/j.eij.2025.100845

Tasneem Duridi , Areej Jaber , Paloma Martínez

Hate speech on social media, particularly in Arabic contexts, poses a serious threat to user well-being and online interaction quality. The complexity of Arabic, with its wide range of dialects, necessitates advanced detection systems to identify harmful communication targeting individuals or groups based on ethnicity, religion, gender, or nationality. To address this challenge, three key Twitter datasets — OSACT-5, LAHS, and arHate — are examined to represent diverse dialects and sociopolitical contexts. These datasets provide valuable resources for investigating the binary classification task of Arabic hate speech detection. Leveraging BERT transformers with their bidirectional contextual understanding enables the capture of nuanced meanings in Arabic expressions, thereby enhancing classification accuracy.

ix prominent Arabic BERT variants — MarBERT, BERT-multilingual, QARiB, CAMeLBERT, AraBERTv0.2-Twitter, and SaudiBERT — are systematically evaluated across the selected datasets. To enhance performance, hyperparameter optimization using Grid Search and Bayesian methods is conducted only on the top three performing models. The optimized models achieve strong results: on OSACT-5, QARiB attains 93.2% accuracy with an F1-score of 82%; on LAHS, QARiB reaches 85.0% accuracy and 87.0% F1; and on arHate, SaudiBERT achieves 94.5% accuracy with a 91.8% F1-score. These results highlight the robustness and adaptability of optimized Arabic BERT models for hate speech detection across diverse dialects and imbalanced datasets, contributing to more reliable moderation of harmful content in Arabic social media.

社交媒体上的仇恨言论，特别是在阿拉伯语境下的仇恨言论，对用户福祉和在线互动质量构成严重威胁。阿拉伯语的复杂性及其广泛的方言，需要先进的检测系统来识别基于种族、宗教、性别或国籍的针对个人或团体的有害通信。为了应对这一挑战，我们研究了三个关键的Twitter数据集——OSACT-5、LAHS和arHate，以代表不同的方言和社会政治背景。这些数据集为研究阿拉伯语仇恨言论检测的二元分类任务提供了宝贵的资源。利用具有双向上下文理解的BERT转换器可以捕获阿拉伯语表达中的细微差别含义，从而提高分类准确性。ix突出的阿拉伯语BERT变体- MarBERT， BERT-multilingual, QARiB, CAMeLBERT, AraBERTv0.2-Twitter和SaudiBERT -在选定的数据集中进行系统评估。为了提高性能，使用网格搜索和贝叶斯方法的超参数优化只在表现最好的三个模型上进行。优化后的模型取得了较好的结果：在OSACT-5上，QARiB的准确率为93.2%，f1得分为82%；在LAHS上，QARiB准确率达到85.0%，F1达到87.0%；在arHate上，SaudiBERT的准确率为94.5%，f1得分为91.8%。这些结果突出了优化的阿拉伯语BERT模型对不同方言和不平衡数据集的仇恨言论检测的鲁棒性和适应性，有助于更可靠地审查阿拉伯社交媒体中的有害内容。

{"title":"Arabic hate speech detection based on BERT models variants","authors":"Tasneem Duridi , Areej Jaber , Paloma Martínez","doi":"10.1016/j.eij.2025.100845","DOIUrl":"10.1016/j.eij.2025.100845","url":null,"abstract":"<div><div>Hate speech on social media, particularly in Arabic contexts, poses a serious threat to user well-being and online interaction quality. The complexity of Arabic, with its wide range of dialects, necessitates advanced detection systems to identify harmful communication targeting individuals or groups based on ethnicity, religion, gender, or nationality. To address this challenge, three key Twitter datasets — OSACT-5, LAHS, and arHate — are examined to represent diverse dialects and sociopolitical contexts. These datasets provide valuable resources for investigating the binary classification task of Arabic hate speech detection. Leveraging BERT transformers with their bidirectional contextual understanding enables the capture of nuanced meanings in Arabic expressions, thereby enhancing classification accuracy.</div><div>ix prominent Arabic BERT variants — MarBERT, BERT-multilingual, QARiB, CAMeLBERT, AraBERTv0.2-Twitter, and SaudiBERT — are systematically evaluated across the selected datasets. To enhance performance, hyperparameter optimization using Grid Search and Bayesian methods is conducted only on the top three performing models. The optimized models achieve strong results: on OSACT-5, QARiB attains 93.2% accuracy with an F1-score of 82%; on LAHS, QARiB reaches 85.0% accuracy and 87.0% F1; and on arHate, SaudiBERT achieves 94.5% accuracy with a 91.8% F1-score. These results highlight the robustness and adaptability of optimized Arabic BERT models for hate speech detection across diverse dialects and imbalanced datasets, contributing to more reliable moderation of harmful content in Arabic social media.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100845"},"PeriodicalIF":4.3,"publicationDate":"2025-11-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578963","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Adaptive Q-Learning-Based Event-Prioritized QoS and incentive optimization for enhancing safety in vehicular fog networks 基于自适应q学习的事件优先QoS和激励优化提高车辆雾网络的安全性

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-19 DOI: 10.1016/j.eij.2025.100821

Sajib Tripura , Qing-Chang Lu , Dhonita Tripura , Md Ibrahim Kholilullah , Arunav Mallik Avi , Mostak Ahamed , Adil Hussain

In the rapidly changing world of Intelligent Transportation Systems (ITS), achieving fast, reliable, and energy-efficient communication in vehicle fog computing (VFC) networks is crucial for safety–critical applications. Current VFC approaches are not apt for safety–critical applications as they are based on static heuristics, QoS focus design which neglects trust, energy and reliability; slow convergence and does not support fairness and responsiveness. Moreover, they do not adaptively prioritize concurrent emergencies, which motivates the development of mobility and criticality-aware adaptive approaches. This study proposes a novel reinforcement learning framework named Q-APERF based on tabular Q-learning agent improved by the Augmented Priority-Entropy Reward Function (APERF). Our approach dynamically adjusts multiple QoS metrics, including latency, reliability, trustworthiness, and energy consumption, while prioritizing overlapping emergencies such as ambulances, crash alerts, and road hazards exponentially. The agent achieves adaptive QoS weighting and discrete vehicular state, and therefore, the message forwarding performance can be enhanced in a highly dynamic environment (i.e., the IoV). Extensive simulations show that it outperforms some of the existing state-of-the-art approaches. The Q-APERF achieves 95.5% of message prioritization accuracy, 75.4% of transmission efficacy in packet loss situation, and 83% of energy efficiency and 80% faster response to emergency events, which illustrates its dynamic resilience adaptability balance QoS and energy consumption perspective. Moreover, we introduce a novel metric, Survival-Weighted Data Integrity (SWDI), to evaluate incentive mechanisms that promote the sustained participation of vehicles and encourage them to share their resources. This holistic view will enable safer and more fault-tolerant smart transportation systems through offering a secure, scalable, and context-aware vehicular communication solution.

在快速变化的智能交通系统（ITS）世界中，在车辆雾计算（VFC）网络中实现快速、可靠和节能的通信对于安全关键应用至关重要。目前的VFC方法不适合安全关键型应用，因为它们基于静态启发式，QoS焦点设计忽略了信任，能量和可靠性；收敛缓慢，不支持公平性和响应性。此外，它们不能自适应地优先处理并发紧急情况，这促使了机动性和临界感知自适应方法的发展。本文提出了一种基于增强型优先熵奖励函数（APERF）改进的表格q -学习代理的强化学习框架Q-APERF。我们的方法动态调整多个QoS指标，包括延迟、可靠性、可信度和能耗，同时优先考虑救护车、碰撞警报和道路危险等重叠紧急情况。agent实现了自适应的QoS权重和离散的车辆状态，因此可以在高度动态的环境（即IoV）中提高消息转发性能。大量的模拟表明，它优于一些现有的最先进的方法。Q-APERF的报文优先级精度达到95.5%，丢包情况下的传输效率达到75.4%，能量效率提高83%，对突发事件的响应速度提高80%，体现了其动态弹性适应性平衡QoS和能耗的视角。此外，我们引入了一种新的度量，生存加权数据完整性（SWDI），以评估促进车辆持续参与和鼓励他们共享资源的激励机制。通过提供安全、可扩展和上下文感知的车辆通信解决方案，这种整体视图将使智能交通系统更安全、更容错。

{"title":"Adaptive Q-Learning-Based Event-Prioritized QoS and incentive optimization for enhancing safety in vehicular fog networks","authors":"Sajib Tripura , Qing-Chang Lu , Dhonita Tripura , Md Ibrahim Kholilullah , Arunav Mallik Avi , Mostak Ahamed , Adil Hussain","doi":"10.1016/j.eij.2025.100821","DOIUrl":"10.1016/j.eij.2025.100821","url":null,"abstract":"<div><div>In the rapidly changing world of Intelligent Transportation Systems (ITS), achieving fast, reliable, and energy-efficient communication in vehicle fog computing (VFC) networks is crucial for safety–critical applications. Current VFC approaches are not apt for safety–critical applications as they are based on static heuristics, QoS focus design which neglects trust, energy and reliability; slow convergence and does not support fairness and responsiveness. Moreover, they do not adaptively prioritize concurrent emergencies, which motivates the development of mobility and criticality-aware adaptive approaches. This study proposes a novel reinforcement learning framework named Q-APERF based on tabular Q-learning agent improved by the Augmented Priority-Entropy Reward Function (APERF). Our approach dynamically adjusts multiple QoS metrics, including latency, reliability, trustworthiness, and energy consumption, while prioritizing overlapping emergencies such as ambulances, crash alerts, and road hazards exponentially. The agent achieves adaptive QoS weighting and discrete vehicular state, and therefore, the message forwarding performance can be enhanced in a highly dynamic environment (i.e., the IoV). Extensive simulations show that it outperforms some of the existing state-of-the-art approaches. The Q-APERF achieves 95.5% of message prioritization accuracy, 75.4% of transmission efficacy in packet loss situation, and 83% of energy efficiency and 80% faster response to emergency events, which illustrates its dynamic resilience adaptability balance QoS and energy consumption perspective. Moreover, we introduce a novel metric, Survival-Weighted Data Integrity (SWDI), to evaluate incentive mechanisms that promote the sustained participation of vehicles and encourage them to share their resources. This holistic view will enable safer and more fault-tolerant smart transportation systems through offering a secure, scalable, and context-aware vehicular communication solution.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100821"},"PeriodicalIF":4.3,"publicationDate":"2025-11-19","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578968","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Design of a federated learning algorithm for power big data privacy computing based on pruning technique and homomorphic encryption 基于剪枝技术和同态加密的电力大数据隐私计算联邦学习算法设计

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-18 DOI: 10.1016/j.eij.2025.100828

Weijian Zhang , Li Di , Jing Zhang , Boyu Liu , Xinyan Wang

This paper proposes a pruned homomorphic encryption-based collaborative federated learning algorithm to address the issues of non-uniform data distribution, overfitting caused by residual gradient features, and privacy leakage risks in power big data computation. The algorithm integrates the advantages of dynamic pruning and homomorphic encryption techniques: pruning mitigates overfitting risks and enhances efficiency by sparsifying the model (reducing the number of parameters by 30% to 50%), while homomorphic encryption ensures the security of gradient aggregation in the ciphertext domain (reducing computation cost by 60%). The specific procedure includes: initializing model parameters and generating encryption keys; designing a federated acceleration algorithm based on dynamic pruning to compress gradients and perform homomorphic encryption; updating model parameters by aggregating ciphertext gradients via weighted averaging; and finally decrypting the aggregated result to obtain the plaintext output. Experimental results demonstrate that the proposed algorithm, combined with automatic meta-pruning technology and a partial homomorphic encryption scheme, achieves a 47% improvement in computational efficiency and a 36% reduction in communication overhead under million-scale data volumes. It supports collaborative training among 50 institutions (with training time per institution for million-level data less than 8 h). By integrating encryption and pruning in a dual mechanism, the algorithm balances privacy protection and model performance, offering a secure and practical privacy-preserving computation solution for the power industry.

针对电力大数据计算中存在的数据分布不均匀、残差梯度特征导致的过拟合、隐私泄露风险等问题，提出了一种基于剪枝同态加密的协同联邦学习算法。该算法综合了动态剪枝和同态加密技术的优点：剪枝通过模型稀疏化降低了过拟合风险，提高了效率（减少了30% ~ 50%的参数数量），而同态加密保证了密文域梯度聚合的安全性（减少了60%的计算成本）。具体过程包括：初始化模型参数，生成加密密钥；设计了一种基于动态剪枝的联邦加速算法来压缩梯度并进行同态加密；采用加权平均法对密文梯度进行聚合，更新模型参数；最后对聚合结果进行解密，得到明文输出。实验结果表明，该算法结合自动元剪枝技术和部分同态加密方案，在百万级数据量下，计算效率提高47%，通信开销降低36%。支持50家机构协同培训（百万级数据每家机构培训时间小于8 h）。该算法将加密和剪枝结合在一个双重机制中，平衡了隐私保护和模型性能，为电力行业提供了一个安全实用的隐私保护计算解决方案。

{"title":"Design of a federated learning algorithm for power big data privacy computing based on pruning technique and homomorphic encryption","authors":"Weijian Zhang , Li Di , Jing Zhang , Boyu Liu , Xinyan Wang","doi":"10.1016/j.eij.2025.100828","DOIUrl":"10.1016/j.eij.2025.100828","url":null,"abstract":"<div><div>This paper proposes a pruned homomorphic encryption-based collaborative federated learning algorithm to address the issues of non-uniform data distribution, overfitting caused by residual gradient features, and privacy leakage risks in power big data computation. The algorithm integrates the advantages of dynamic pruning and homomorphic encryption techniques: pruning mitigates overfitting risks and enhances efficiency by sparsifying the model (reducing the number of parameters by 30% to 50%), while homomorphic encryption ensures the security of gradient aggregation in the ciphertext domain (reducing computation cost by 60%). The specific procedure includes: initializing model parameters and generating encryption keys; designing a federated acceleration algorithm based on dynamic pruning to compress gradients and perform homomorphic encryption; updating model parameters by aggregating ciphertext gradients via weighted averaging; and finally decrypting the aggregated result to obtain the plaintext output. Experimental results demonstrate that the proposed algorithm, combined with automatic <em>meta</em>-pruning technology and a partial homomorphic encryption scheme, achieves a 47% improvement in computational efficiency and a 36% reduction in communication overhead under million-scale data volumes. It supports collaborative training among 50 institutions (with training time per institution for million-level data less than 8 h). By integrating encryption and pruning in a dual mechanism, the algorithm balances privacy protection and model performance, offering a secure and practical privacy-preserving computation solution for the power industry.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100828"},"PeriodicalIF":4.3,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578966","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Hair removal and lesion segmentation of dermoscopic images for classification of skin cancer using deep neural networks 基于深度神经网络的皮肤镜图像脱毛和病灶分割用于皮肤癌分类

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-18 DOI: 10.1016/j.eij.2025.100844

Aqib Shahzaib , Abdul Basit Siddiqui , Nadeem Anjum , Masood Ur Rehman , Naeem Ramzan

Skin cancer is a prevalent health issue worldwide. Therefore, early detection through automated deep learning systems is crucial for saving lives. Hair presence in dermoscopic image presents diagnostic challenges by obscuring lesion features and complicating analysis due to variations in hair characteristics (density, color, and distribution), which can lead to diagnostic errors. In this work, a new and comprehensive approach is introduced to enhance the automatic classification of skin lesions. We propose an Efficient Hair Removal (EHR) technique that combines a Deep Residual U-Net with the TELEA inpainting algorithm, effectively eliminating hair artifacts from dermoscopic images. For precise lesion delineation, a Deep Residual U-Net model for skin lesion segmentation is also employed. The ISIC2019 dataset is used for skin lesion classification. Our approach progresses through five experimental stages, each building upon the previous. Starting with dataset balancing, which improved classification accuracy by 5%, we then applied our EHR framework, further boosting accuracy by 2.53%. The integration of skin lesion segmentation contributed to an additional 1.5% improvement. In the last, we use modified DenseNet169 architecture, which achieves a top accuracy of 97.74% on the ISIC2019 dataset, outperforming existing techniques. For lesion segmentation, Deep Residual U-Net achieved good results on the ISIC2018 dataset, with an Intersection over Union (IoU) of 0.8981 and a Dice Similarity Score (DSC) of 0.946.

皮肤癌是世界范围内普遍存在的健康问题。因此，通过自动深度学习系统进行早期检测对于挽救生命至关重要。皮肤镜图像中的毛发存在给诊断带来了挑战，因为毛发特征（密度、颜色和分布）的变化模糊了病变特征，使分析变得复杂，这可能导致诊断错误。在这项工作中，提出了一种新的、综合的方法来增强皮肤损伤的自动分类。我们提出了一种高效脱毛（EHR）技术，该技术结合了深度残留U-Net和TELEA染色算法，有效地消除了皮肤镜图像中的毛发伪影。为了精确描绘病变，还采用了一种用于皮肤病变分割的Deep Residual U-Net模型。ISIC2019数据集用于皮肤病变分类。我们的方法经历了五个实验阶段，每个阶段都建立在前一个阶段的基础上。从数据集平衡开始，将分类准确率提高了5%，然后应用我们的EHR框架，进一步将准确率提高了2.53%。整合皮肤病变分割有助于额外1.5%的改善。最后，我们使用改进的DenseNet169架构，该架构在ISIC2019数据集上达到97.74%的最高准确率，优于现有技术。对于病灶分割，Deep Residual U-Net在ISIC2018数据集上取得了很好的效果，交集比（Intersection over Union, IoU）为0.8981,Dice Similarity Score （DSC）为0.946。

{"title":"Hair removal and lesion segmentation of dermoscopic images for classification of skin cancer using deep neural networks","authors":"Aqib Shahzaib , Abdul Basit Siddiqui , Nadeem Anjum , Masood Ur Rehman , Naeem Ramzan","doi":"10.1016/j.eij.2025.100844","DOIUrl":"10.1016/j.eij.2025.100844","url":null,"abstract":"<div><div>Skin cancer is a prevalent health issue worldwide. Therefore, early detection through automated deep learning systems is crucial for saving lives. Hair presence in dermoscopic image presents diagnostic challenges by obscuring lesion features and complicating analysis due to variations in hair characteristics (density, color, and distribution), which can lead to diagnostic errors. In this work, a new and comprehensive approach is introduced to enhance the automatic classification of skin lesions. We propose an Efficient Hair Removal (EHR) technique that combines a Deep Residual U-Net with the TELEA inpainting algorithm, effectively eliminating hair artifacts from dermoscopic images. For precise lesion delineation, a Deep Residual U-Net model for skin lesion segmentation is also employed. The ISIC2019 dataset is used for skin lesion classification. Our approach progresses through five experimental stages, each building upon the previous. Starting with dataset balancing, which improved classification accuracy by 5%, we then applied our EHR framework, further boosting accuracy by 2.53%. The integration of skin lesion segmentation contributed to an additional 1.5% improvement. In the last, we use modified DenseNet169 architecture, which achieves a top accuracy of 97.74% on the ISIC2019 dataset, outperforming existing techniques. For lesion segmentation, Deep Residual U-Net achieved good results on the ISIC2018 dataset, with an Intersection over Union (IoU) of 0.8981 and a Dice Similarity Score (DSC) of 0.946.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100844"},"PeriodicalIF":4.3,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578970","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Optimizing language model fine-tuning and quantization for enhanced medical question answering 优化语言模型微调和量化，增强医疗问题回答

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-18 DOI: 10.1016/j.eij.2025.100836

D. Somashekhara Reddy , Ravindra Raman Cholla , G. Nagendra Babu , T.R. Mahesh , Anu Sayal

This paper introduces an end-to-end fine-tuning and compression solution for the EleutherAI GPT-Neo 125 M language model. It employs a new optimization procedure that boosts the performance of the model in the task of medical question answering. The model was seeded with a corpus from the MedQuAD dataset, which consists of 47,457 question–answer pairs retrieved from authorized NIH websites. The MedLine Plus data were, however, excluded due to licensing restrictions. The corpus covers a large range of medical subjects, such as diseases and medicines, and includes different question types and rich annotations for enhancing information retrieval (IR) and natural language processing (NLP) tasks. Fine-tuning was conducted for 3750 training steps, wherein the loss during training continuously dropped from 3.0094 to 0.2343. This shows that the model successfully learned and adapted during the process. The enhanced model incorporated LoRA and 4-bit quantization techniques, specifically intended to improve computational efficiency and model scalability. The measures of performance were evaluated by measuring the BLEU, ROUGE, and TER scores, which were utilized to compare the original and optimized model. The optimized model exhibited considerable improvements in performance metrics. The BLEU value rose from 0.0489 to 0.0706, the ROUGE value rose from 0.0621 to 0.1996, and the TER value fell from 0.9053 to 0.8120. These findings justify the efficacy of fine-tuning and quantization interventions. The results indicate that targeted model optimization can enhance the accuracy and dependability of AI-driven medical question-answering systems quite significantly.

本文介绍了一种针对EleutherAI GPT-Neo 125 M语言模型的端到端微调和压缩解决方案。它采用了一种新的优化过程，提高了模型在医疗问答任务中的性能。该模型使用MedQuAD数据集的语料库作为种子，该数据集由从NIH授权网站检索的47,457对问答组成。然而，由于许可限制，MedLine Plus数据被排除在外。该语料库涵盖了广泛的医学主题，如疾病和药物，并包括不同的问题类型和丰富的注释，以增强信息检索（IR）和自然语言处理（NLP）任务。对3750个训练步骤进行微调，训练过程中的损失从3.0094持续下降到0.2343。这表明该模型在此过程中成功地进行了学习和适应。增强模型结合了LoRA和4位量化技术，专门用于提高计算效率和模型可扩展性。通过测量BLEU， ROUGE和TER评分来评估性能指标，用于比较原始模型和优化模型。优化后的模型在性能指标上有了相当大的改进。BLEU值从0.0489上升到0.0706，ROUGE值从0.0621上升到0.1996，TER值从0.9053下降到0.8120。这些发现证明了微调和量化干预的有效性。结果表明，有针对性的模型优化可以显著提高人工智能驱动的医疗问答系统的准确性和可靠性。

{"title":"Optimizing language model fine-tuning and quantization for enhanced medical question answering","authors":"D. Somashekhara Reddy , Ravindra Raman Cholla , G. Nagendra Babu , T.R. Mahesh , Anu Sayal","doi":"10.1016/j.eij.2025.100836","DOIUrl":"10.1016/j.eij.2025.100836","url":null,"abstract":"<div><div>This paper introduces an end-to-end fine-tuning and compression solution for the EleutherAI GPT-Neo 125 M language model. It employs a new optimization procedure that boosts the performance of the model in the task of medical question answering. The model was seeded with a corpus from the MedQuAD dataset, which consists of 47,457 question–answer pairs retrieved from authorized NIH websites. The MedLine Plus data were, however, excluded due to licensing restrictions. The corpus covers a large range of medical subjects, such as diseases and medicines, and includes different question types and rich annotations for enhancing information retrieval (IR) and natural language processing (NLP) tasks. Fine-tuning was conducted for 3750 training steps, wherein the loss during training continuously dropped from 3.0094 to 0.2343. This shows that the model successfully learned and adapted during the process. The enhanced model incorporated LoRA and 4-bit quantization techniques, specifically intended to improve computational efficiency and model scalability. The measures of performance were evaluated by measuring the BLEU, ROUGE, and TER scores, which were utilized to compare the original and optimized model. The optimized model exhibited considerable improvements in performance metrics. The BLEU value rose from 0.0489 to 0.0706, the ROUGE value rose from 0.0621 to 0.1996, and the TER value fell from 0.9053 to 0.8120. These findings justify the efficacy of fine-tuning and quantization interventions. The results indicate that targeted model optimization can enhance the accuracy and dependability of AI-driven medical question-answering systems quite significantly.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100836"},"PeriodicalIF":4.3,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578965","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Ant Colony Optimization for feature selection in breast cancer classification 基于蚁群算法的乳腺癌分类特征选择

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-18 DOI: 10.1016/j.eij.2025.100847

Ashokkumar P. , Sateesh Kumar TV , Mudassir Khan , Mazliham Mohd Su’ud , Muhammad Mansoor Alam , Saurav Mallik

Breast cancer remains one of the world’s most prevalent cancers that mostly affects women. Recent advancements in machine learning enable early detection of breast cancer with high accuracy, significantly reducing the mortality risk. Feature selection plays a crucial role in enhancing the performance of machine learning classification models by reducing dimensions, optimizing classification outcomes, and improving computational efficiency. In this study, we propose a nature-inspired feature selection method using Ant Colony Optimization (ACO) to enhance the classification of breast cancer. We have utilized soft voting of five machine learning models: k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) to evaluate the proposed methods. The experimental results on the two benchmark datasets, Wisconsin Breast Cancer Database (WBCD) and Wisconsin Diagnostic Breast Cancer (WDBC) demonstrates the effectiveness of the proposed approach, achieving classification accuracies of 99.79 and 99.71, respectively, outperforming the existing state-of-the-art methods.

乳腺癌仍然是世界上最普遍的癌症之一，主要影响女性。机器学习的最新进展使乳腺癌的早期检测具有很高的准确性，大大降低了死亡风险。特征选择通过降维、优化分类结果和提高计算效率，对提高机器学习分类模型的性能起着至关重要的作用。在这项研究中，我们提出了一种基于蚁群优化（Ant Colony Optimization， ACO）的自然特征选择方法来增强乳腺癌的分类。我们使用了五种机器学习模型的软投票：k-近邻（kNN）、支持向量机（SVM）、逻辑回归（LR）、决策树（DT）和随机森林（RF）来评估所提出的方法。在Wisconsin Breast Cancer Database （WBCD）和Wisconsin Diagnostic Breast Cancer （WDBC）两个基准数据集上的实验结果证明了该方法的有效性，分类准确率分别达到99.79和99.71，优于现有的最先进的方法。

{"title":"Ant Colony Optimization for feature selection in breast cancer classification","authors":"Ashokkumar P. , Sateesh Kumar TV , Mudassir Khan , Mazliham Mohd Su’ud , Muhammad Mansoor Alam , Saurav Mallik","doi":"10.1016/j.eij.2025.100847","DOIUrl":"10.1016/j.eij.2025.100847","url":null,"abstract":"<div><div>Breast cancer remains one of the world’s most prevalent cancers that mostly affects women. Recent advancements in machine learning enable early detection of breast cancer with high accuracy, significantly reducing the mortality risk. Feature selection plays a crucial role in enhancing the performance of machine learning classification models by reducing dimensions, optimizing classification outcomes, and improving computational efficiency. In this study, we propose a nature-inspired feature selection method using Ant Colony Optimization (ACO) to enhance the classification of breast cancer. We have utilized soft voting of five machine learning models: k-Nearest Neighbors (kNN), Support Vector Machine (SVM), Logistic Regression (LR), Decision Tree (DT), and Random Forest (RF) to evaluate the proposed methods. The experimental results on the two benchmark datasets, Wisconsin Breast Cancer Database (WBCD) and Wisconsin Diagnostic Breast Cancer (WDBC) demonstrates the effectiveness of the proposed approach, achieving classification accuracies of 99.79 and 99.71, respectively, outperforming the existing state-of-the-art methods.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100847"},"PeriodicalIF":4.3,"publicationDate":"2025-11-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578967","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Optimization of logistics business process decision model based on multi-objective optimization algorithm 基于多目标优化算法的物流业务流程决策模型优化

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-17 DOI: 10.1016/j.eij.2025.100835

Junfeng Fan , Jingheng Wang

Logistics is becoming harder due to several difficulties. Globalization, changing client needs, and complex supply networks are examples. Traditional planning often disappoints. The inability to be adaptable, flexible, and rapid in finding effective solutions is the biggest issue. This research found that adaptive algorithms and artificial intelligence allow the Logistic Intelligence Multiple-Objective Decisions Model (LIMOD) to optimize complicated logistical business processes. After that, the study results were revealed. Several competing objectives are considered when designing the LIMOD. Goals are to shorten time to delivery, save money, and maximize resource use. LIMOD’s real-time priority adjusting technology is advanced. This mechanism represents a major LIMOD advancement. The concept’s dynamic adaptability makes it valuable in many logistical situations, especially those including supply chain uncertainty. This is particularly true for ambiguous supply networks. Production, online retail, and transportation companies with multi-layer supply chains could profit from LIMOD. It can connect without affecting business applications and be customized for a domain. Logistics managers can maximize decision possibilities and trade-offs with LIMOD, improving strategic planning. This improves talents. This study advances intelligent, sustainable logistics optimization systems. Businesses can confidently navigate global supply networks with this technology. Experiments showed LIMOD’s better performance. These tests used various datasets and logistical scenarios. Addressing three or more objective simultaneously, LIMOD provides 18% greater solution quality, 27% quicker completion times, and 23% greater effectiveness than typical optimization processes. Compared to other optimization methods. Learn about LIMOD here.

由于一些困难，物流变得越来越困难。全球化、不断变化的客户需求和复杂的供应网络就是例子。传统的计划常常令人失望。最大的问题是不能适应、灵活和快速地找到有效的解决方案。本研究发现，自适应算法和人工智能允许物流智能多目标决策模型（LIMOD）优化复杂的物流业务流程。之后，研究结果就出来了。在设计LIMOD时要考虑几个相互竞争的目标。目标是缩短交付时间，节省资金，并最大限度地利用资源。LIMOD的实时优先级调整技术是先进的。这种机制代表了LIMOD的一个主要进步。该概念的动态适应性使其在许多物流情况下具有价值，特别是在包括供应链不确定性的情况下。对于模棱两可的供应网络尤其如此。拥有多层供应链的生产、在线零售和运输公司可以从LIMOD中获利。它可以在不影响业务应用程序的情况下进行连接，并且可以针对某个域进行定制。物流经理可以最大化决策的可能性和权衡与LIMOD，改进战略规划。这提高了天赋。本研究提出了智能、可持续的物流优化系统。借助这项技术，企业可以自信地驾驭全球供应网络。实验表明LIMOD具有较好的性能。这些测试使用了各种数据集和逻辑场景。LIMOD可以同时处理三个或更多的目标，与典型的优化过程相比，解决方案质量提高18%，完成时间缩短27%，效率提高23%。与其他优化方法相比。在这里了解LIMOD。

{"title":"Optimization of logistics business process decision model based on multi-objective optimization algorithm","authors":"Junfeng Fan , Jingheng Wang","doi":"10.1016/j.eij.2025.100835","DOIUrl":"10.1016/j.eij.2025.100835","url":null,"abstract":"<div><div>Logistics is becoming harder due to several difficulties. Globalization, changing client needs, and complex supply networks are examples. Traditional planning often disappoints. The inability to be adaptable, flexible, and rapid in finding effective solutions is the biggest issue. This research found that adaptive algorithms and artificial intelligence allow the Logistic Intelligence Multiple-Objective Decisions Model (LIMOD) to optimize complicated logistical business processes. After that, the study results were revealed. Several competing objectives are considered when designing the LIMOD. Goals are to shorten time to delivery, save money, and maximize resource use. LIMOD’s real-time priority adjusting technology is advanced. This mechanism represents a major LIMOD advancement. The concept’s dynamic adaptability makes it valuable in many logistical situations, especially those including supply chain uncertainty. This is particularly true for ambiguous supply networks. Production, online retail, and transportation companies with multi-layer supply chains could profit from LIMOD. It can connect without affecting business applications and be customized for a domain. Logistics managers can maximize decision possibilities and trade-offs with LIMOD, improving strategic planning. This improves talents. This study advances intelligent, sustainable logistics optimization systems. Businesses can confidently navigate global supply networks with this technology. Experiments showed LIMOD’s better performance. These tests used various datasets and logistical scenarios. Addressing three or more objective simultaneously, LIMOD provides 18% greater solution quality, 27% quicker completion times, and 23% greater effectiveness than typical optimization processes. Compared to other optimization methods. Learn about LIMOD here.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100835"},"PeriodicalIF":4.3,"publicationDate":"2025-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578876","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Golden Jackal Driven Optimization for a transparent and interpretable Intrusion Detection System using explainable AI to revolutionize cybersecurity 金豺驱动的优化透明和可解释的入侵检测系统，使用可解释的人工智能来彻底改变网络安全

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-15 DOI: 10.1016/j.eij.2025.100837

Syed Haider Ali Shah , Lalarukh Haseeb Akhtar , Muhammad Nadeem Ali , Byung-Seo Kim

The increasing connectivity of today’s digital world and contents necessitate robust security solutions to address the escalating range of cyber threats as well as Copyright infringement. Intrusion Detection Systems (IDS) have emerged as essential tools for managing, analyzing, and generating cybersecurity responses against potential cyberattacks. However, IDS face persistent challenges, including detection performance, deployment efficiency, and reliability in real-time scenarios, which remain active areas of research. In addition to these challenges, a novel issue has recently gained attention: the lack of explainability and transparency in IDS predictions. This limitation significantly affects security practitioners’ confidence in system reliability and restricts the practical utilization of the insights produced. To address these concerns, this paper proposes a joint approach to enhance both the performance and explainability of IDS by integrating the Golden Jackal Optimization (GJO) algorithm for cyberattack detection. Furthermore, we incorporate Explainable Artificial Intelligence (XAI) to provide a clear and comprehensive interpretation of the model’s predictions. Notably, the proposed XAI-based model achieved an impressive accuracy of 99.82% and a miss rate of just 0.19%, underscoring its efficiency, trustworthiness, transparency, and interpretability key attributes essential for human operators managing intelligent cybersecurity systems.

当今数字世界和内容的连通性日益增强，需要强大的安全解决方案来应对不断升级的网络威胁和版权侵权。入侵检测系统（IDS）已经成为管理、分析和生成针对潜在网络攻击的网络安全响应的基本工具。然而，IDS面临着持续的挑战，包括检测性能、部署效率和实时场景下的可靠性，这些仍然是研究的活跃领域。除了这些挑战之外，最近还出现了一个新的问题：IDS预测缺乏可解释性和透明度。这一限制极大地影响了安全从业者对系统可靠性的信心，并限制了所产生的见解的实际利用。为了解决这些问题，本文提出了一种联合方法，通过集成用于网络攻击检测的Golden Jackal Optimization （GJO）算法来提高IDS的性能和可解释性。此外，我们结合了可解释人工智能（XAI）来提供对模型预测的清晰和全面的解释。值得注意的是，提出的基于xai的模型达到了令人印象深刻的99.82%的准确率和0.19%的缺失率，强调了其效率、可信度、透明度和可解释性等关键属性，这些属性对于人类操作员管理智能网络安全系统至关重要。

{"title":"Golden Jackal Driven Optimization for a transparent and interpretable Intrusion Detection System using explainable AI to revolutionize cybersecurity","authors":"Syed Haider Ali Shah , Lalarukh Haseeb Akhtar , Muhammad Nadeem Ali , Byung-Seo Kim","doi":"10.1016/j.eij.2025.100837","DOIUrl":"10.1016/j.eij.2025.100837","url":null,"abstract":"<div><div>The increasing connectivity of today’s digital world and contents necessitate robust security solutions to address the escalating range of cyber threats as well as Copyright infringement. Intrusion Detection Systems (IDS) have emerged as essential tools for managing, analyzing, and generating cybersecurity responses against potential cyberattacks. However, IDS face persistent challenges, including detection performance, deployment efficiency, and reliability in real-time scenarios, which remain active areas of research. In addition to these challenges, a novel issue has recently gained attention: the lack of explainability and transparency in IDS predictions. This limitation significantly affects security practitioners’ confidence in system reliability and restricts the practical utilization of the insights produced. To address these concerns, this paper proposes a joint approach to enhance both the performance and explainability of IDS by integrating the Golden Jackal Optimization (GJO) algorithm for cyberattack detection. Furthermore, we incorporate Explainable Artificial Intelligence (XAI) to provide a clear and comprehensive interpretation of the model’s predictions. Notably, the proposed XAI-based model achieved an impressive accuracy of 99.82% and a miss rate of just 0.19%, underscoring its efficiency, trustworthiness, transparency, and interpretability key attributes essential for human operators managing intelligent cybersecurity systems.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100837"},"PeriodicalIF":4.3,"publicationDate":"2025-11-15","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145578971","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Advanced hybrid UNet architectures for retinal vessel segmentation 视网膜血管分割的先进混合UNet架构

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-14 DOI: 10.1016/j.eij.2025.100823

Naif Alsharabi , Amira Echtioui , Abdulaziz Alayba , Gharbi Alshammari , Habib Hamam

Retinal blood vessel segmentation is a critical step in the early detection and diagnosis of various vision-threatening diseases, including diabetic retinopathy, hypertension, and glaucoma. Manual segmentation by medical professionals is time-consuming, subjective, and prone to variability, highlighting the need for robust automated solutions. Recent advancements in deep learning have shown significant promise in addressing these challenges by enabling accurate and efficient segmentation of retinal blood vessels from fundus images. In this paper, we propose three advanced deep learning architectures for retinal blood vessel segmentation: VGG16-UNet, VGG19-UNet, and ResNet50-UNet. These models combine the strengths of pre-trained convolutional neural networks (CNNs) with the U-Net architecture, leveraging transfer learning to enhance feature extraction and segmentation performance. We evaluate the proposed models on a publicly available retinal image dataset, achieving Dice coefficients of 79%, 79% and 80% for VGG16-UNet, VGG19-UNet, and ResNet50-UNet, respectively. Our results demonstrate that the ResNet50-UNet model outperforms the other two variants and surpasses several state-of-the-art methods in terms of segmentation accuracy and robustness. This study underscores the potential of deep learning-based approaches for improving the automation and reliability of retinal blood vessel segmentation, paving the way for more efficient and accurate diagnostic tools in ophthalmology.

视网膜血管分割是早期发现和诊断各种视力威胁疾病的关键步骤，包括糖尿病视网膜病变、高血压和青光眼。由医疗专业人员进行的手动分割耗时、主观且容易发生变化，这突出了对健壮的自动化解决方案的需求。深度学习的最新进展表明，通过从眼底图像中准确有效地分割视网膜血管，可以解决这些挑战。在本文中，我们提出了三种先进的视网膜血管分割深度学习架构：VGG16-UNet， VGG19-UNet和ResNet50-UNet。这些模型结合了预训练卷积神经网络（cnn）和U-Net架构的优势，利用迁移学习来增强特征提取和分割性能。我们在一个公开可用的视网膜图像数据集上评估了所提出的模型，VGG16-UNet、VGG19-UNet和ResNet50-UNet的Dice系数分别为79%、79%和80%。我们的结果表明，ResNet50-UNet模型优于其他两种变体，并且在分割精度和鲁棒性方面超过了几种最先进的方法。这项研究强调了基于深度学习的方法在提高视网膜血管分割的自动化和可靠性方面的潜力，为眼科中更有效和准确的诊断工具铺平了道路。

{"title":"Advanced hybrid UNet architectures for retinal vessel segmentation","authors":"Naif Alsharabi , Amira Echtioui , Abdulaziz Alayba , Gharbi Alshammari , Habib Hamam","doi":"10.1016/j.eij.2025.100823","DOIUrl":"10.1016/j.eij.2025.100823","url":null,"abstract":"<div><div>Retinal blood vessel segmentation is a critical step in the early detection and diagnosis of various vision-threatening diseases, including diabetic retinopathy, hypertension, and glaucoma. Manual segmentation by medical professionals is time-consuming, subjective, and prone to variability, highlighting the need for robust automated solutions. Recent advancements in deep learning have shown significant promise in addressing these challenges by enabling accurate and efficient segmentation of retinal blood vessels from fundus images. In this paper, we propose three advanced deep learning architectures for retinal blood vessel segmentation: VGG16-UNet, VGG19-UNet, and ResNet50-UNet. These models combine the strengths of pre-trained convolutional neural networks (CNNs) with the U-Net architecture, leveraging transfer learning to enhance feature extraction and segmentation performance. We evaluate the proposed models on a publicly available retinal image dataset, achieving Dice coefficients of 79%, 79% and 80% for VGG16-UNet, VGG19-UNet, and ResNet50-UNet, respectively. Our results demonstrate that the ResNet50-UNet model outperforms the other two variants and surpasses several state-of-the-art methods in terms of segmentation accuracy and robustness. This study underscores the potential of deep learning-based approaches for improving the automation and reliability of retinal blood vessel segmentation, paving the way for more efficient and accurate diagnostic tools in ophthalmology.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100823"},"PeriodicalIF":4.3,"publicationDate":"2025-11-14","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528260","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Fake news detection on social media using triple-attention mechanism optimized by advanced tailor optimization algorithm 基于高级定制优化算法优化的三注意机制的社交媒体假新闻检测

IF 4.3 3区计算机科学 Q1 COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE

Egyptian Informatics Journal

Pub Date : 2025-11-13 DOI: 10.1016/j.eij.2025.100815

Huiru Jin , Peng Wang

Problem: While the phenomenal expansion of social media is reshaping the ways through which information is disseminated, such platforms also liberally facilitate the rampant propagation of fake news, which has serious implications on public opinion, democratic processes, and social well-being. Traditional fake news detection techniques as rule-based systems and conventional machine learning models are poles apart from the speed, variety, and virtual processing of social media content and mostly are characterized by low accuracy, poor generalization, and inability to capture multimodal cues. Method: In order to accommodate this contention, a new deep learning mechanism based on the Triple-attention Mechanism optimized by the Advanced Tailor Optimization Algorithm is proposed in this study. Triple-attention Mechanism perceives, estimates, and considers all three most important dimensions for generating news content-the semantic textual part, contextual features, and user activities-generating more all-embracing and context-aware analysis. Advanced Tailor Optimization Algorithm alters dynamically the learning rate and transmission other hyperparameters at the training phase, thereby enhancing convergence, stability, and generalization of the model from adaptive exploration and exploitation. Results: The model is evaluated on four benchmark datasets-FakeNewsNet, LIAR, PolitiFact, and GossipCop-using extensive performance metrics. The experimental findings outdid those of other evaluations, achieving a rarity of accuracy values of 93, 88, 91, and 90; F1-scores of 93, 88, 91, and 90; specificity values of 95, 92, 93, and 91; and sensitivity scores of 94, 89, 92, and 91. This consistent performance also excelled against six cutting-edge baseline models, with respect to all metrics, including precision, Kappa score, Matthew’s correlation coefficient, Dice Similarity Coefficient, and Intersection over Union. Conclusion: These findings establish the fact that the TAM-ATOA framework is efficient and robust in detecting fake news across various domains. The resulting framework, which integrates multi-dimensional attentions with a new adaptive optimization strategy, develops a solution that is more accurate, scalable, and reliable in the battle against misinformation in real-life social media environments.

问题：虽然社交媒体的惊人扩张正在重塑信息传播的方式，但这些平台也为假新闻的猖獗传播提供了自由，这对公众舆论、民主进程和社会福祉产生了严重影响。传统的假新闻检测技术，如基于规则的系统和传统的机器学习模型，与社交媒体内容的速度、多样性和虚拟处理截然不同，它们大多具有准确性低、泛化能力差、无法捕捉多模态线索的特点。方法：为了适应这一争论，本研究提出了一种基于高级定制优化算法优化的三注意机制的新的深度学习机制。三重注意机制感知、估计并考虑生成新闻内容的所有三个最重要的维度——语义文本部分、上下文特征和用户活动——从而生成更全面和上下文感知的分析。高级定制优化算法在训练阶段动态改变学习率和传输等超参数，从而增强了自适应探索和开发模型的收敛性、稳定性和泛化性。结果：该模型在四个基准数据集（fakenewsnet、LIAR、PolitiFact和gossip cop）上进行了评估，并使用了广泛的性能指标。实验结果优于其他评估，达到了93,88,91和90的稀有性准确性值；f1得分为93、88、91和90分；特异性值分别为95、92、93和91；敏感度得分分别是94 89 92 91。这种一致的性能也优于六个尖端的基线模型，涉及所有指标，包括精度，Kappa分数，马修相关系数，骰子相似系数和交集。结论：这些发现证明了TAM-ATOA框架在检测不同领域的假新闻方面是高效和稳健的。由此产生的框架将多维关注与新的自适应优化策略相结合，开发了一种更准确、可扩展、更可靠的解决方案，可以在现实社会媒体环境中与错误信息作斗争。

{"title":"Fake news detection on social media using triple-attention mechanism optimized by advanced tailor optimization algorithm","authors":"Huiru Jin , Peng Wang","doi":"10.1016/j.eij.2025.100815","DOIUrl":"10.1016/j.eij.2025.100815","url":null,"abstract":"<div><div>Problem: While the phenomenal expansion of social media is reshaping the ways through which information is disseminated, such platforms also liberally facilitate the rampant propagation of fake news, which has serious implications on public opinion, democratic processes, and social well-being. Traditional fake news detection techniques as rule-based systems and conventional machine learning models are poles apart from the speed, variety, and virtual processing of social media content and mostly are characterized by low accuracy, poor generalization, and inability to capture multimodal cues. Method: In order to accommodate this contention, a new deep learning mechanism based on the Triple-attention Mechanism optimized by the Advanced Tailor Optimization Algorithm is proposed in this study. Triple-attention Mechanism perceives, estimates, and considers all three most important dimensions for generating news content-the semantic textual part, contextual features, and user activities-generating more all-embracing and context-aware analysis. Advanced Tailor Optimization Algorithm alters dynamically the learning rate and transmission other hyperparameters at the training phase, thereby enhancing convergence, stability, and generalization of the model from adaptive exploration and exploitation. Results: The model is evaluated on four benchmark datasets-FakeNewsNet, LIAR, PolitiFact, and GossipCop-using extensive performance metrics. The experimental findings outdid those of other evaluations, achieving a rarity of accuracy values of 93, 88, 91, and 90; F1-scores of 93, 88, 91, and 90; specificity values of 95, 92, 93, and 91; and sensitivity scores of 94, 89, 92, and 91. This consistent performance also excelled against six cutting-edge baseline models, with respect to all metrics, including precision, Kappa score, Matthew’s correlation coefficient, Dice Similarity Coefficient, and Intersection over Union. Conclusion: These findings establish the fact that the TAM-ATOA framework is efficient and robust in detecting fake news across various domains. The resulting framework, which integrates multi-dimensional attentions with a new adaptive optimization strategy, develops a solution that is more accurate, scalable, and reliable in the battle against misinformation in real-life social media environments.</div></div>","PeriodicalId":56010,"journal":{"name":"Egyptian Informatics Journal","volume":"32 ","pages":"Article 100815"},"PeriodicalIF":4.3,"publicationDate":"2025-11-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528272","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":3,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0