|
Abstract
|
This research investigates modern sentiment mining methods, particularly for languages like Nepali, Arabic, and Turkish, which present unique linguistic challenges. It proposes a transformer-based framework using pre-trained language models like BERT to analyze Twitter data sentiment, addressing noise and short-text issues common in social media. The methodology involves data collection, preprocessing, model fine-tuning, and evaluation, with the model outperforming baseline models like BiLSTM and CNN-LSTM. The study highlights the advantages of transformers, including contextual embeddings, parallel processing, and transfer learning, and contributes a novel approach that leverages attention mechanisms for accurate sentiment classification in short-text data, while also improving time and memory efficiency compared to hybrid models.
|