<![CDATA[The rapid growth of social media has generated massive and complex amounts of data, creating the need for more effective analytical methods. Deep learning has emerged as one of the approaches capable of automatically identifying patterns within such data; however, systematic studies that comprehensively map its development across social media platforms remain limited. This study aims to analyze research trends in deep learning on social media platforms during the period of 2021–2026 using the Systematic Literature Review (SLR) approach. The analysis focuses on annual publication trends, the most frequently studied social media platforms, the deep learning models employed, dominant research topics, and future research opportunities. Data were collected through a literature search of scientific articles in the Google Scholar and Garuda databases based on predefined inclusion and exclusion criteria, resulting in nine eligible articles for analysis. The findings indicate that research on the application of deep learning in social media has shown significant advancements in model architectures, ranging from single models such as Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU) to hybrid models and Transformer-based architectures, including BERT-LSTM and IndoBERT. These approaches have been predominantly applied to sentimen analysis, fake news detection, and hate speech detection. The findings further reveal that CNN remained the most widely used model during the 2021–2026 period, although a gradual shift toward Transformer-based models such as BERT and IndoBERT has begun to emerge. This trend highlights the potential of Transformer-based architectures to improve semantic context understanding and multimedia data processing in future research.]]>
Copyrights © 2026
