Topic detection and short-text analysis have been significantly transformed by integrating machine learning (ML) techniques and large language models (LLMs) such as BERT and GPT, particularly in platforms like Twitter. These advanced models outperform traditional rule-based and statistical approaches by leveraging transformer architectures and semantic embedding techniques (e.g., word embeddings) to uncover text's latent themes and contextual relationships. Even in low-resource language settings, LLMs can capture semantic nuances and support robust text classification and dynamic topic modeling. However, Arabic-language applications face unique challenges, primarily due to the scarcity of high-quality, task-specific annotated datasets, especially for domains like synthetic content identification and fake news detection. Successful model training in Arabic requires extensive corpora, careful linguistic preprocessing, and sensitivity to morphological complexity and dialectal variability. Additionally, LLMs are limited by computational limitations related to input length, which restricts the capacity for scaling when working with large volumes of text. In conclusion, future research should focus on establishing hybrid frameworks with contextual fine-tuning for domains, cross-lingual transfer learning, and better management of computational memory to address these obstacles and completely tap into the possibilities of ML-driven text analytics in resource-constrained settings.
Dawood et al. (Fri,) studied this question.