A Literature Review on Stock Price Prediction by Integrating Social Media Sentiment and Market Indicators Using Deep Learning Approaches

This paper reviews how social media and news sentiment can be integrated with market indicators to improve stock prediction, with emphasis on deep learning and multimodal fusion. A systematic review is conducted on literature from 2010 to 2024 across IEEE Xplore, ACM Digital Library, Scopus, and arXiv using explicit inclusion/exclusion criteria and a PRISMA-style screening procedure; foundational methods predating this window are also referenced where appropriate. To organize diverse findings, a four-dimensional taxonomy is proposed covering data sources, sentiment construction methods, fusion strategies, and predictive architectures. A comparative synthesis of representative studies yields three consistent patterns: multimodal approaches outperform unimodal (sentiment-only or price-only) baselines; hybrid fusion provides the most reliable balance between robustness and cross-modal interaction; and architectures coupling finance-specific encoders (e.g., FinBERT) with sequence models (e.g., LSTM or transformer-based forecasters) achieve strong error reductions, particularly in high-volatility regimes. Limitations persist—noisy and asynchronously timed sentiment signals, heterogeneous performance across markets and regimes, inconsistent evaluation protocols that hinder comparability, and complex yet opaque models that challenge adoption. In response, a practical evaluation checklist is provided and concrete future directions are outlined, including attention-based alignment, domain-specific pretraining across languages/markets, reinforcement learning for decision-making, cross-market robustness testing, and explainability.

Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.