Hybrid Deep Learning-Based Sentiment Analysis for Stock Market Euphoria: A Review of Recent Advances

This systematic literature review explores the detection of stock market euphoria by synthesizing recent advancements in hybrid deep learning and sentiment analysis. Traditional financial models often fail to capture market euphoria, an extreme form of investor sentiment that can lead to speculative bubbles. Consequently, advanced computational approaches that integrate behavioral finance with machine learning have become necessary. This review follows the PRISMA 2020 guidelines to identify, evaluate, and synthesize recent studies focusing on high-frequency social media data and sophisticated natural language processing techniques, particularly Transformer-based models (e.g., BERT) and recurrent neural networks (e.g., LSTM). Our synthesis reveals a predominant focus on general sentiment classification (positive/negative) within established, English-speaking markets. Key findings indicate that hybrid architectures that leverage BERT’s contextual understanding and LSTM’s temporal modeling consistently outperform standalone models. However, we identify significant research gaps, including a lack of studies dedicated to characterizing the specific emotional state of “euphoria” and the limited application of languagespecific models in emerging markets such as Indonesia. Key challenges include model complexity, data synchronization, and high computational costs. This review highlights a clear opportunity for future research to develop culturally and linguistically tailored hybrid models, such as combining IndoBERT with LSTM, which offers significant research potential.

Authors: