Cross-Domain Sentiment Analysis using Transfer Learning: A Literature Review on Natural Language Model Adaptation from Social-Media to Macroeconomic Indicator Prediction

Munawaroh, Silvi

International Journal of Research and Applied Technology (INJURATECH)
Vol. 5 No. 2 (2025): December 2025

This study reviews the efficacy of transfer learning in adapting sentiment analysis from social media domains to macroeconomic indicator prediction. The study evaluates existing literature on natural language model architectures, specifically Transformer-based models, performing domain adaptation from informal social media discourse to formal economic contexts. Findings indicate that pre-trained models significantly enhance predictive accuracy for data-scarce economic indicators by capturing real-time public perception. While effective in addressing labeled data sparsity, primary challenges involve linguistic noise and inherent demographic biases within social media datasets. Transfer learning serves as a critical bridge in transforming public sentiment into predictive economic signals. This cross-domain approach provides a dynamic, supplementary instrument for policymakers to monitor macroeconomic fluctuations through digital behavioral patterns.