Cross-domain sentiment analysis (CDSA) is a Natural Language Processing (NLP) task that involves analyzing and predicting sentiment across different domains or areas, such as product reviews, social media, and movie feedback. In CDSA, models trained on one domain (e.g., electronics) must adapt to ...

Cross-domain sentiment analysis (CDSA) is a Natural Language Processing (NLP) task that involves analyzing and predicting sentiment across different domains or areas, such as product reviews, social media, and movie feedback. In CDSA, models trained on one domain (e.g., electronics) must adapt to a different domain (e.g., books) where language, vocabulary, and contextual cues can vary widely. Traditional sentiment analysis models perform well within specific domains but often struggle with new, unseen domains, limiting their effectiveness in real-world applications like social media monitoring, customer feedback analysis, and market research. We propose a novel approach combining Elmo’s deep contextualized embeddings with BERT's robust sentiment classification. ELMo captures word-level dependencies and nuances, while BERT, with its transformer architecture, excels in understanding complex language structures. This is the first approach to use ELMo for cross-domain sentiment analysis in Bangla, addressing a critical gap in this language. Experimental results show that our ELMo-BERT model outperforms traditional models, achieving higher accuracy and robustness, making it highly adaptable for real-world applications across diverse Bangla datasets.

In this study we analyzed different well-established machine learning (ML) and deep learning (DL) supervised models to enable the risk prediction of maternal health, thus offering a viable and systematic technique to automatically identify pregnancy risk. The Maternal Health Risk Data Set, which covers ...

In this study we analyzed different well-established machine learning (ML) and deep learning (DL) supervised models to enable the risk prediction of maternal health, thus offering a viable and systematic technique to automatically identify pregnancy risk. The Maternal Health Risk Data Set, which covers various critical attributes such as age, blood pressure, blood sugar, body temperature, heart rate, and risk level, was applied [8]. Data pretreatment methods, including deleting missing data (if any) and conducting feature scaling and selection, were incorporated to create the model. Different ML models were created and tested, including but not limited to Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB), as well as deep learning architectures such as Convolutional Neural Network (CNN), Recurrent Neural Network (RNN), and Long Short-Term Memory (LSTM). Model performances were evaluated using metric measures, including accuracy and F1 scores. Of them, CNN showed the highest accuracy (98.58), exceeding alternative models by its capacity to uncover spatial correlations crucial for successful risk prediction. CNN shows great accuracy, which indicates that its real-life clinical application in predicting highrisk pregnancies would result in a considerable improvement in maternal care. Adding AI-driven models to existing healthcare settings could assist in the faster and more accurate evaluation of pregnancy risk, particularly in low-resource settings, boosting focused preventative therapy and evidence-based clinical decision-making. The expanding presence of AI has the potential to revolutionize healthcare, taking us closer to scalable automated solutions for maternal health that correspond with global healthcare development goals, with implications from this study

Breast cancer is one of the most prevalent and life-threatening diseases world- wide, emphasizing the need for accurate and early diagnosis. This research explores the potential of machine learning (ML) and deep learning (DL) techniques for...

Breast cancer is one of the most prevalent and life-threatening diseases world- wide, emphasizing the need for accurate and early diagnosis. This research explores the potential of machine learning (ML) and deep learning (DL) techniques for breast cancer prediction, leveraging their ability to process and analyze complex medical data. A comprehensive dataset was used to evaluate and compare multiple models, including traditional ML algorithms such as Support Vector Machines and Random Forests, alongside advanced DL architectures like Convolutional Neural Networks (CNNs). The study focuses on feature selection, data preprocessing, and hyperparameter opti- mization to enhance model performance. Results demonstrate that the DL models outperform traditional ML approaches in accuracy, sensitivity, and specificity. This paper provides insights into the capabilities and limitations of these methods, highlighting the role of artificial intelligence in improving clinical decision-making and patient outcomes.

Suicide is an undeniable threat to our society nowadays. People of all ages are committing suicide. Most of them remain in depression before committing suicide. They often share their posts and thoughts on social networks to express the suffering they are going through. Suicide prevention has become a critical concern in the digital age, with social media platforms providing valuable information on people's mental health. This study ...

Suicide is an undeniable threat to our society nowadays. People of all ages are committing suicide. Most of them remain in depression before committing suicide. They often share their posts and thoughts on social networks to express the suffering they are going through. Suicide prevention has become a critical concern in the digital age, with social media platforms providing valuable information on people's mental health. This study will classify suicide-related posts using machine learning and with the help of different deep neural network models, for example, BERT, CNN, ANN, LSTM, and a hybrid CNN-LSTM approach. We used Twitter and Reddit datasets, which have two columns: tweet and the target column (potential suicide or non-suicide), applying Natural Language Processing (NLP) techniques for preprocessing and feature extraction. Our research systematically evaluates multiple models to determine their effectiveness in predicting the ideation of suicide. This research comes up with the idea that deep learning-based models, especially BERT, CNN, and CNN-LSTM, outperform traditional ML classifiers for predicting the presence of tendencies with suicide, with 98.12\% accuracy in BERT rather than 94\% in SVM and Random Forest. This study shows that AI-powered methods can be used to rapidly assess the odds of suicide risk in early stages. It contributes to the ever-growing scope of AI in mental state by showing the application of NLP methodologies to the monitoring and intervention of suicide risk in real time.

Artificial intelligence (AI) technology has been applied more and more in healthcare applications for decision support, the lack of transparency behind many AI systems often results in diminished trust and suboptimal adoption. In the human computer interaction (HCI) domain, it is important to know how the design of interfaces affects user's perception of AI-powered systems for healthcare. This paper examines the influence of explainable artificial intelligence (XAI) on user trust, usability, and confidence in a healthcare decision support interface. We implemented ...

Artificial intelligence (AI) technology has been applied more and more in healthcare applications for decision support, the lack of transparency behind many AI systems often results in diminished trust and suboptimal adoption. In the human - computer interaction (HCI) domain, it is important to know how the design of interfaces affects user's perception of AI-powered systems for healthcare. This paper examines the influence of explainable artificial intelligence (XAI) on user trust, usability, and confidence in a healthcare decision support interface. We implemented a prototype of the diabetes risk prediction based in machine learning and work with two interface versions: black-box, which displays only the prognostic output information; explainable, which also returns human understandable justifications for those predictions. A controlled user study was performed to compare user reactions between the two interfaces with personalized usability and trust measures. Experimental results demonstrate that the explainable interface has a clear and significant improvement on perceived trust, usability, and user confidence over the black-box one. Our work demonstrates the significance of interpretability in healthcare AI systems, and offers actionable design recommendations on developing accountable and user-centered AI-enabled applications for health concerns. This paper offers evidence for the emerging area of human - AI interactions in healthcare.

Fake news detection is an important task in modern information systems, due to the high propagation speed of misinformation on online platforms. In this study, a graph- based deep learning framework is proposed to detect fake news from textual news data. The dataset is preprocessed and feature representations are formed using RoBERTa embeddings, a graph structure is built in which news articles are considered nodes whereas semantic relationships provide the edges. The findings are ...

Fake news detection is an important task in modern information systems, due to the high propagation speed of misinformation on online platforms. In this study, a graph- based deep learning framework is proposed to detect fake news from textual news data. The dataset is preprocessed and feature representations are formed using RoBERTa embeddings, a graph structure is built in which news articles are considered nodes whereas semantic relationships provide the edges. The findings are demonstrated by applying a Graph Neural Network (GNN) to learn contextual node representations and classify articles as authentic or counterfeit. Its accuracy is 97%, precision 98%, recall 97% and F1-score is equal to 97% capturing the structural dependency of news samples. The method is investigated in terms of confusion matrix, ROC - AUC, Precision - Recall curves as well as t-SNE visualization showing the effectiveness and reliability of the approach and demonstrating how graph-based learning is suitable for scalable fake news detection. This work presents a reproducible pipeline for graph-based text classifiers, and highlights the potential application of GNNs in misinformation prevention.