Market Trend Discovery Through Deep Learning-Driven Twitter Sentiment for Stock Forecasting

Short-range stock price forecasting presents considerable challenges due to the multifaceted and rapidly evolving nature of market signals. ‎Twitter, as a social media platform, delivers immediate public sentiment, serving as a valuable complement to traditional price-based metrics. This research introduces an operational framework merging Twitter sentiment classification with time-series forecasting that integrates ‎both sentiment and technical market features. The sentiment analysis component employs an ensemble approach combining LSTM, CNN, ‎and fine-tuned BERT architectures. These sentiment indicators feed into a neural network designed for time-series analysis to forecast directional price movements. Testing on datasets from January 2020 through December 2022 yielded 87.3% accuracy in sentiment classification ‎and 82.1% accuracy in directional prediction. These outcomes indicate that well-processed social media sentiment enhances short-term trading signals when combined with conventional technical indicators.