Analyzing Sentiments in eLearning: A Comparative Study of Bangla and Romanized Bangla Text Using Transformers

IEEE Access, Год журнала: 2024, Номер 12, С. 89144 - 89162

Опубликована: Янв. 1, 2024

In the modern world, learning is becoming increasingly critical due to rapid technological breakthroughs, which highlight need for continuous skill development in both personal and professional spheres. As a result, eLearning cutting-edge approach education that delivers lessons, courses, instructional materials remotely via digital technology Internet. It makes more flexible accessible by enabling users interact with teachers online access classes or other content. Sentiment analysis an technique evaluates user opinions, typically written feedback, improve overall quality of instruction course. e-learning feedback has been extensively studied several languages, except Bangla Romanized Bangla. The three datasets produced were one Bangla, combination Romanized. Three contained 3178 3090 6268 texts. divided into categories: positive, negative, neutral. validation was conducted using Krippendorff's alpha Cohen's kappa metrics, ensuring reliability consistency dataset annotations. Several techniques used train preprocessed datasets, including transformers, deep learning, machine ensemble hybrid approaches. Transformer-based algorithms, such as XLM-RoBERTa, outperformed others terms accuracy, achieving highest values 89.46% 85.81% Combined datasets. At 89.59%, ANN demonstrated exceptional performance on dataset.

Язык: Английский

---

Bias and Cyberbullying Detection and Data Generation Using Transformer Artificial Intelligence Models and Top Large Language Models

Electronics, Год журнала: 2024, Номер 13(17), С. 3431 - 3431

Опубликована: Авг. 29, 2024

Despite significant advancements in Artificial Intelligence (AI) and Large Language Models (LLMs), detecting mitigating bias remains a critical challenge, particularly on social media platforms like X (formerly Twitter), to address the prevalent cyberbullying these platforms. This research investigates effectiveness of leading LLMs generating synthetic biased data evaluates proficiency transformer AI models within both authentic contexts. The study involves semantic analysis feature engineering dataset over 48,000 sentences related collected from Twitter (before it became X). Utilizing state-of-the-art tools such as ChatGPT-4, Pi AI, Claude 3 Opus, Gemini-1.5, biased, cyberbullying, neutral were generated deepen understanding human-generated data. including DeBERTa, Longformer, BigBird, HateBERT, MobileBERT, DistilBERT, BERT, RoBERTa, ELECTRA, XLNet initially trained classify subsequently fine-tuned, optimized, experimentally quantized. focuses intersectional multilabel classification detect cyberbullying. Additionally, proposes two prototype applications: one that detects using an approach innovative CyberBulliedBiasedBot combines generation detection content.

Язык: Английский

---

Hybrid Deep Transfer Learning Framework for Humerus Fracture Detection and Classification from X-ray Images

Опубликована: Июнь 21, 2024

Язык: Английский

---

Enhancing Cybersecurity: Hybrid Deep Learning Approaches to Smishing Attack Detection

Systems, Год журнала: 2024, Номер 12(11), С. 490 - 490

Опубликована: Ноя. 14, 2024

Smishing attacks, a sophisticated form of cybersecurity threats conducted via Short Message Service (SMS), have escalated in complexity with the widespread adoption mobile devices, making it increasingly challenging for individuals to distinguish between legitimate and malicious messages. Traditional phishing detection methods, such as feature-based, rule-based, heuristic, blacklist approaches, struggled keep pace rapidly evolving tactics employed by attackers. To enhance address these challenges, this paper proposes hybrid deep learning approach that combines Bidirectional Gated Recurrent Units (Bi-GRUs) Convolutional Neural Networks (CNNs), referred CNN-Bi-GRU, accurate identification classification smishing attacks. The SMS Phishing Collection dataset was used, preparatory procedure involving transformation unstructured text data into numerical representations training Word2Vec on preprocessed text. Experimental results demonstrate proposed CNN-Bi-GRU model outperforms existing achieving an overall highest accuracy 99.82% detecting This study provides empirical analysis effectiveness techniques detection, offering more precise efficient solution communications.

Язык: Английский

---

Investigating the Effectiveness of Deep Learning and Machine Learning for Bangla Poems Genre Classification

Опубликована: Июнь 21, 2024

Язык: Английский

---

Empowering Digital Civility with an NLP Approach for Detecting Twitter Cyberbullying through Boosted Ensembles

ACM Transactions on Asian and Low-Resource Language Information Processing, Год журнала: 2024, Номер 23(12), С. 1 - 31

Опубликована: Окт. 7, 2024

As the number of social networking sites grows, so do cyber dangers. Cyberbullying is harmful behavior that uses technology to intimidate, harass, or harm someone, often on media platforms like 𝕏 (formerly known as Twitter). Machine learning optimal approach for cyberbullying detection process large amounts data, identify patterns offensive behavior, and automate corpus tweets. To threats using a trained model, boosted ensemble (BE) technique assessed with various machine algorithms such convolutional neural network (CNN), long short-term memory (LSTM), naive Bayes (NB), decision tree (DT), support vector (SVM), bidirectional LSTM (BILSTM), recurrent (RNN-LSTM), multi-modal (MMCD), random forest (RF). These classifiers are vectorized data classify tweets threats. The proposed framework can detect cases precisely significance work lies in detecting mitigating real time, it impacts enhancing safety well-being users by reducing instances other comparative analysis done metrics accuracy, precision, recall, F1-score, comparison results show BE outperforms compared its overall performance. Respectively, accuracy rates CNN, LSTM, NB, DT, SVM, RF, BILSTM, 92.5%, 93.5%, 84.6%, 88%, 89.3%, 92%, 93.75%, 96%; precision RNN-LSTM, 90.2%, 91.3%, 85%, 86%, 91.6%, 92.1%, 94%; recall 89.8%, 90.7%, 90%, 82%, 88.67%, 89%, 91.04%, 93.7%; F1-scores MMCD, 90.6%, 91.8%, 84.56% 87.2%, 94.89%.

Язык: Английский

---

Hyperparameter Tuning of Pre-Trained Architectures for Multi-Modal Cyberbullying Detection

Advances in information security, privacy, and ethics book series, Год журнала: 2024, Номер unknown, С. 465 - 502

Опубликована: Окт. 4, 2024

Over the past decade, cyberbullying has become a pervasive issue, particularly among young individuals, causing growing concern within society. The rise of social media provided fertile ground for incidents to occur. In this work, proposed deep learning based Multi-Modal Cyberbullying Detection(MMC) technique identify on both text and image data combination. This MMC involves two pre trained architectures generate feature vector representations. RoBERTa Xception are employed extract features from respectively. LightGBM classifier is used classify multi-modal bullying or non-bullying . hyperparameter tuning applied improve classification performance detection data. experiments conducted 2100 samples combined image. efficiently classifies with f1-score 80% outperforms as compared existing approaches.

Язык: Английский

---

Detecting and Understanding Cyberbullying in Bengali: An Explainable AI Approach

Опубликована: Сен. 25, 2024

Язык: Английский

---

Enhancing Cyberbullying Detection on Social Media Using Transformer Models

Опубликована: Июнь 19, 2024

Язык: Английский

---