Shakil Hossain
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Shakil Hossain has a total of 20 co-authors. Here are the top co-authors by number of shared publications:
- Mithun Hossain
Department of Computer Science and Engineering, Bangladesh University of Business and Technology, Dhaka, Bangladesh
6 shared publications - Muhammad Firoz Mridha
Department of Computer Science and Engineering, American International University-Bangladesh, Dhaka, Bangladesh
5 shared publications - Sudipto Chaki
Department of Computer Science and Engineering, Bangladesh University of Business and Technology, Dhaka, Bangladesh
5 shared publications - Mejdl Safran
Department of Computer Science, College of Computer and Information Sciences, King Saud University, Riyadh 11543, Saudi Arabia
2 shared publications - Saifur Rahman
Department of Computer Science and Engineering, Bangladesh University of Business & Technology, Dhaka, Bangladesh
2 shared publications - Sanjara
Department of Computer Science and Engineering, Bangladesh University of Business and Technology (BUBT), Dhaka, Bangladesh
2 shared publications
Latest Publications
- Open Access
- 18 March 2026
- Published by Elsevier in SSRN Electronic Journal
Reasoning under Ambiguity: Uncertainty-Aware Multilingual Emotion Classification under Partial Supervision
Contemporary knowledge-based systems increasingly rely on multilingual emotion identification to support intelligent decision-making, yet they face major challenges due to emotional ambiguity and incomplete supervision. Emotion recognition from text is inherently uncertain because multiple emotional states often co-occur and emotion annotations are frequently missing or heterogeneous. Most existing multi-label emotion classification methods assume fully observed labels and rely on deterministic learning objectives, which can lead to biased learning and unreliable predictions under partial supervision. This paper introduces Reasoning under Ambiguity, an uncertainty-aware framework for multilingual multi-label emotion classification that explicitly aligns learning with annotation uncertainty. The proposed approach uses a shared multilingual encoder with language-specific optimization and an entropy-based ambiguity weighting mechanism that down-weights highly ambiguous training instances rather than treating missing labels as negative evidence. A mask-aware objective with positive-unlabeled regularization is further incorporated to enable robust learning under partial supervision. Experiments on English, Spanish, and Arabic emotion classification benchmarks demonstrate consistent improvements over strong baselines across multiple evaluation metrics, along with improved training stability, robustness to annotation sparsity, and enhanced interpretability.
- 18 March 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessReasoning under Ambiguity: Uncertainty-Aware Multilingual Emotion Classification under Partial Supervision
Contemporary knowledge-based systems increasingly rely on multilingual emotion identification to support intelligent decision-making, yet they face major challenges due to emotional ambiguity and incomplete supervision. Emotion recognition from text is inherently uncertain because multiple emotional states often co-occur and emotion annotations are frequently missing or heterogeneous. Most existing multi-label emotion classification methods assume fully observed labels and rely on deterministic learning objectives, which can lead to biased learning and unreliable predictions under partial supervision. This paper introduces Reasoning under Ambiguity, an uncertainty-aware framework for multilingual multi-label emotion classification that explicitly aligns learning with annotation uncertainty. The proposed approach uses a shared multilingual encoder with language-specific optimization and an entropy-based ambiguity weighting mechanism that down-weights highly ambiguous training instances rather than treating missing labels as negative evidence. A mask-aware objective with positive-unlabeled regularization is further incorporated to enable robust learning under partial supervision. Experiments on English, Spanish, and Arabic emotion classification benchmarks demonstrate consistent improvements over strong baselines across multiple evaluation metrics, along with improved training stability, robustness to annotation sparsity, and enhanced interpretability.
- 20 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 7 (5) , 2726-2738
Co-AttenDWG: Coattentive Dimension-Wise Gating and Expert Fusion for Multimodal Offensive Content Detection
Multi-modal learning has emerged as a crucial research direction, as integrating textual and visual information can substantially enhance performance in tasks such as classification, retrieval, and scene understanding. Despite advances with large pre-trained models, existing approaches often suffer from insufficient cross-modal interactions and rigid fusion strategies, failing to fully harness the complementary strengths of different modalities. To address these limitations, we propose Co-AttenDWG, co-attention with dimension-wise gating, and expert fusion. Our approach first projects textual and visual features into a shared embedding space, where a dedicated co-attention mechanism enables simultaneous, fine-grained interactions between modalities. This is further strengthened by a dimension-wise gating network, which adaptively modulates feature contributions at the channel level to emphasize salient information. In parallel, dual-path encoders independently refine modality-specific representations, while an additional cross-attention layer aligns the modalities further. The resulting features are aggregated via an expert fusion module that integrates learned gating and self-attention, yielding a robust unified representation. Experimental results on the MIMIC and SemEval Memotion 1.0 datasets show that Co-AttenDWG achieves state-of-the-art performance and superior cross-modal alignment, highlighting its effectiveness for diverse multi-modal applications.
- 20 October 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Transactions on Artificial Intelligence
- Vol. 7 (5) , 2726-2738
Co-AttenDWG: Coattentive Dimension-Wise Gating and Expert Fusion for Multimodal Offensive Content Detection
Multi-modal learning has emerged as a crucial research direction, as integrating textual and visual information can substantially enhance performance in tasks such as classification, retrieval, and scene understanding. Despite advances with large pre-trained models, existing approaches often suffer from insufficient cross-modal interactions and rigid fusion strategies, failing to fully harness the complementary strengths of different modalities. To address these limitations, we propose Co-AttenDWG, co-attention with dimension-wise gating, and expert fusion. Our approach first projects textual and visual features into a shared embedding space, where a dedicated co-attention mechanism enables simultaneous, fine-grained interactions between modalities. This is further strengthened by a dimension-wise gating network, which adaptively modulates feature contributions at the channel level to emphasize salient information. In parallel, dual-path encoders independently refine modality-specific representations, while an additional cross-attention layer aligns the modalities further. The resulting features are aggregated via an expert fusion module that integrates learned gating and self-attention, yielding a robust unified representation. Experimental results on the MIMIC and SemEval Memotion 1.0 datasets show that Co-AttenDWG achieves state-of-the-art performance and superior cross-modal alignment, highlighting its effectiveness for diverse multi-modal applications.
- 25 September 2025
- Published by Wiley in Journal of Cutaneous Pathology
- Vol. 52 (12) , 765-781
Multi‐Scale Attention Fusion With Depthwise Separable Convolutions for Efficient Skin Cancer Detection
Skin cancer is a major global health concern, where early and accurate detection is crucial for improving patient outcomes. Traditional diagnostic methods, such as manual visual inspection and conventional machine learning models, often suffer from subjectivity, high computational costs, and limited annotated data. Althoug deep learning has improved automated skin cancer detection, existing models face challenges like overfitting, insufficient generalization, and complex architectures that limit real‐time clinical application. To address these limitations, we propose MAF‐DermNet, a deep learning framework that integrates Multi‐Scale Attention Fusion (MAF) with depthwise separable convolutions for efficient and accurate skin cancer detection. Our approach enhances data diversity using DCGAN‐based synthetic augmentation to improve model robustness. By leveraging multi‐resolution inputs and a residual attention block, MAF‐DermNet effectively captures subtle lesion features while preserving critical low‐level information. Extensive experiments demonstrate exceptional performance, with accuracy exceeding 99.9% and macro F1 scores above 99.5%. In addition to its superior classification capabilities, MAF‐DermNet offers enhanced interpretability and computational efficiency, making it well‐suited for clinical deployment. Future work will focus on integrating clinical metadata and optimizing the model for diverse healthcare settings to further improve early diagnosis and treatment outcomes.
- 25 September 2025
- Published by Wiley in Journal of Cutaneous Pathology
- Vol. 52 (12) , 765-781
Multi‐Scale Attention Fusion With Depthwise Separable Convolutions for Efficient Skin Cancer Detection
Skin cancer is a major global health concern, where early and accurate detection is crucial for improving patient outcomes. Traditional diagnostic methods, such as manual visual inspection and conventional machine learning models, often suffer from subjectivity, high computational costs, and limited annotated data. Althoug deep learning has improved automated skin cancer detection, existing models face challenges like overfitting, insufficient generalization, and complex architectures that limit real‐time clinical application. To address these limitations, we propose MAF‐DermNet, a deep learning framework that integrates Multi‐Scale Attention Fusion (MAF) with depthwise separable convolutions for efficient and accurate skin cancer detection. Our approach enhances data diversity using DCGAN‐based synthetic augmentation to improve model robustness. By leveraging multi‐resolution inputs and a residual attention block, MAF‐DermNet effectively captures subtle lesion features while preserving critical low‐level information. Extensive experiments demonstrate exceptional performance, with accuracy exceeding 99.9% and macro F1 scores above 99.5%. In addition to its superior classification capabilities, MAF‐DermNet offers enhanced interpretability and computational efficiency, making it well‐suited for clinical deployment. Future work will focus on integrating clinical metadata and optimizing the model for diverse healthcare settings to further improve early diagnosis and treatment outcomes.
- 12 September 2025
- Published by Elsevier in Neural Networks
- Vol. 194, 108089
A vision-language model for multitask classification of memes
- 12 September 2025
- Published by Elsevier in Neural Networks
- Vol. 194, 108089
A vision-language model for multitask classification of memes
- 5 September 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 329
Enhancing sentiment analysis with local and global memory in heterogeneous graph neural networks
- 5 September 2025
- Published by Elsevier in Knowledge-Based Systems
- Vol. 329
Enhancing sentiment analysis with local and global memory in heterogeneous graph neural networks
- Open Access
- 9 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 85428-85446
Revolutionizing Mental Health Sentiment Analysis With BERT-Fuse: A Hybrid Deep Learning Model
Mental health disorders profoundly affect intellectual ability, emotional resilience, and social stability, leading to lasting effects on both individuals and society. Increasing global concern over mental health calls for innovative solutions, with sentiment analysis emerging as a promising approach for early diagnosis and intervention. Traditional methods are limited by the complexity of mental health-related texts, which contain specialized terminology and domain-specific nuances. To address these challenges, we propose BERT-Fuse, a hybrid deep learning model combining traditional feature extraction techniques (TF-IDF and Bag-of-Words) with advanced neural architectures, including Bidirectional Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Transformer blocks, and Convolutional Neural Networks (CNN). We evaluated three word embedding techniques-BERT, GloVe, and FastText within BERT-Fuse and found that BERT embeddings achieve the highest performance in terms of accuracy, precision, recall, and F1-score. The model was trained on a balanced dataset of text samples representing seven mental health conditions. An ablation study highlights the contributions of key model components to its performance. BERT-Fuse shows promise as a scalable, high-precision system for mental health detection with an impressive average test accuracy of 97.05%. Our model bridges the gap between deep learning and psychological text analysis, paving the way for future mental health assessment tools that are more intelligent and easily accessible.
- 9 May 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 13, 85428-85446
Open AccessRevolutionizing Mental Health Sentiment Analysis With BERT-Fuse: A Hybrid Deep Learning Model
Mental health disorders profoundly affect intellectual ability, emotional resilience, and social stability, leading to lasting effects on both individuals and society. Increasing global concern over mental health calls for innovative solutions, with sentiment analysis emerging as a promising approach for early diagnosis and intervention. Traditional methods are limited by the complexity of mental health-related texts, which contain specialized terminology and domain-specific nuances. To address these challenges, we propose BERT-Fuse, a hybrid deep learning model combining traditional feature extraction techniques (TF-IDF and Bag-of-Words) with advanced neural architectures, including Bidirectional Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Transformer blocks, and Convolutional Neural Networks (CNN). We evaluated three word embedding techniques-BERT, GloVe, and FastText within BERT-Fuse and found that BERT embeddings achieve the highest performance in terms of accuracy, precision, recall, and F1-score. The model was trained on a balanced dataset of text samples representing seven mental health conditions. An ablation study highlights the contributions of key model components to its performance. BERT-Fuse shows promise as a scalable, high-precision system for mental health detection with an impressive average test accuracy of 97.05%. Our model bridges the gap between deep learning and psychological text analysis, paving the way for future mental health assessment tools that are more intelligent and easily accessible.
- Open Access
- 8 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 1979-1987
Dimension-Wise Gated Cross-Attention for Multimodal Sentiment Analysis
- 8 May 2025
- Published by Association for Computing Machinery (ACM)
- p. 1979-1987
Open AccessDimension-Wise Gated Cross-Attention for Multimodal Sentiment Analysis
- Open Access
- 24 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 198046-198066
A Hybrid Attention-Based Transformer Model for Arabic News Classification Using Text Embedding and Deep Learning
Efficient classification of Arabic news items has become more crucial for efficient information management and analysis due to the fast growth of online news material. This paper proposes a hybrid Attention-Based Transformer Model (ABTM) for Arabic news categorization that uses deep learning and classical text representations to improve classification accuracy and interpretability. Given the increasing amount of Arabic news materials, robust categorization systems are crucial for properly managing and analyzing this information. To deal with the complexities of the Arabic language and enrich the dataset, we used a thorough preparation pipeline that includes text cleaning, tokenization, lemmatization, and data augmentation approaches. We combined a bespoke attention embedder with classic TF-IDF and Bag-of-Words features to provide a comprehensive feature set that includes both the text’s contextual and statistical aspects. We benchmarked our technique using cutting-edge Arabic language models, such as AraBERTv1-base and asafaya/bert-base-arabic. We use (local interpretable model agnostic explanation) text explainer to offer insights into model predictions, improving our findings’ interpretability. Our results show that the ABTM strategy considerably enhances classification performance, with high accuracy and reasonable explanations for model decisions. This classification includes a wide range of news categories, including politics, sports, culture, the economy, and a variety of themes, representing the diversity of Arabic news. This study contributes to the field of Arabic natural language processing by offering a novel method that combines deep learning with traditional techniques, thereby advancing the state of Arabic news classification. Enhanced classification accuracy and interpretability facilitate better management and understanding of the rich and growing Arabic news content, supporting informed decision-making and knowledge discovery.
- 24 December 2024
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Access
- Vol. 12, 198046-198066
Open AccessA Hybrid Attention-Based Transformer Model for Arabic News Classification Using Text Embedding and Deep Learning
Efficient classification of Arabic news items has become more crucial for efficient information management and analysis due to the fast growth of online news material. This paper proposes a hybrid Attention-Based Transformer Model (ABTM) for Arabic news categorization that uses deep learning and classical text representations to improve classification accuracy and interpretability. Given the increasing amount of Arabic news materials, robust categorization systems are crucial for properly managing and analyzing this information. To deal with the complexities of the Arabic language and enrich the dataset, we used a thorough preparation pipeline that includes text cleaning, tokenization, lemmatization, and data augmentation approaches. We combined a bespoke attention embedder with classic TF-IDF and Bag-of-Words features to provide a comprehensive feature set that includes both the text’s contextual and statistical aspects. We benchmarked our technique using cutting-edge Arabic language models, such as AraBERTv1-base and asafaya/bert-base-arabic. We use (local interpretable model agnostic explanation) text explainer to offer insights into model predictions, improving our findings’ interpretability. Our results show that the ABTM strategy considerably enhances classification performance, with high accuracy and reasonable explanations for model decisions. This classification includes a wide range of news categories, including politics, sports, culture, the economy, and a variety of themes, representing the diversity of Arabic news. This study contributes to the field of Arabic natural language processing by offering a novel method that combines deep learning with traditional techniques, thereby advancing the state of Arabic news classification. Enhanced classification accuracy and interpretability facilitate better management and understanding of the rich and growing Arabic news content, supporting informed decision-making and knowledge discovery.