Embedding and Topic Modeling Techniques for Short Text Analysis on Social Media: A Systematic Literature Review

Introduction: The analysis of short texts from social media is critical for gaining insights but is challenged by data sparsity and noise. Integrating embedding and topic modeling techniques has emerged as a key solution. Methods: This study conducted a Systematic Literature Review (SLR) following PRISMA guidelines. A systematic search across IEEE, ScienceDirect, and Scopus databases was performed to identify relevant studies, which were then screened and selected based on predefined inclusion and exclusion criteria. Results: The analysis of 22 included studies revealed a clear methodological trend toward hybrid models that integrate transformer-based embeddings, such as BERT, with topic modeling frameworks. These integrated approaches consistently demonstrated superior performance in generating coherent topics and improving downstream task accuracy compared to standalone or traditional methods. However, limitations related to model generalizability, computational cost, and domain adaptation were identified. Conclusions: The integration of contextual embeddings with topic models is the most effective approach for short-text analysis on social media. Future research should focus on developing more adaptive and efficient models, including fine-tuning language models on domain-specific corpora and exploring the integration of Large Language Models (LLMs) to enhance automation and accuracy.