Stock Price Prediction Accuracy Utilizing Social Media Sentiment

Anotacija

This paper examines the influence of social media sentiment from Twitter/X on Tesla (TSLA) stock price direction prediction. An automated pipeline was implemented using Apache Airflow, combining tweet scraping, FinBERT-based sentiment extraction, and a dual-task LSTM model performing simultaneous directional classification and magnitude regression. Over 278,000 hyperparameter configurations were tested. The best model achieved a weighted F1 score of 0.706 during training, yet live paper trading simulation over 63 trading days yielded only 46% accuracy. Post features were also explored and included in the prediction. The results suggest that raw daily sentiment aggregation from Twitter alone is insufficient for reliable stock price movement prediction.