A SYSTEMATIC REVIEW OF DEEP LEARNING AND SENTIMENT-DRIVEN MODELS FOR CRYPTOCURRENCY PRICE PREDICTION

Abstract

The very high volatility and fast evolution of cryptocurrency markets have revealed the weaknesses of conventional price-based forecasting models, and more behaviorally informed and adaptive predictive models are required. This paper is a systematic review, which aims to assess the connection between deep learning and sentiment based models and their improvement of cryptocurrency price prediction through the integration of investor psychology, market stories and nonlinear dynamics. The research will attempt to synthesize the evidence on the effectiveness of models, current trends in methodology, and applicability to the real world based on fragmented previous research. The systematic review of the literature was done according to PRISMA guidelines. The predefined inclusion and exclusion criteria were used to retrieve peer-reviewed journal articles via Scopus, targeting the articles that were published between 2019 and 2024. It is a review of the empirical and conceptual literature utilizing deep learning architectures (e.g. LSTM, CNN, GRU) and methods of sentiment analysis (e.g. transformer-based models and topic modelling e.g. BERTopic). The results indicate that sentimental deep learning models are reported to be more effective than traditional statistical and machine-learning models in the case of predicting the cryptocurrency prices, especially when the market is volatile. Sentiment activity based on social media and news information are very useful in generating more accurate forecasts, volatility alerts and risk sensitivity in various studies. This paper provides a systematic thematic research synthesis of sentiment-based deep learning studies that have been conducted in the cryptocurrency markets, along with theoretical, methodological, and practical contributions. It offers a basis to the further studies in the field of explainable AI, cross-market validation, and responsible application of predictive analytics in digital finance.

Keywords: Cryptocurrency price prediction; Deep learning; Sentiment analysis; BERTopic; Behavioral finance; Systematic review.