Analyzing Stock Market Trends using Social Media user Moods and Social Influence using Lightweight BERT-BiGRU-A Model

Predicting fluctuations in financial time series, such as the S&P 500, is crucial for making informed investment decisions. Recent studies suggest that machine learning methods can effectively identify non-linear relationships within stock market data. However, the non-stationary nature and severe volatility of financial markets pose significant challenges for accurate trend forecasting. This study investigates the utilization of current transaction data to predict future movements in financial time series. Proper sequencing of feature selection, preprocessing, and model training is emphasized to ensure robust performance. An emerging feature selection strategy, L1-LR, is explored alongside the BERT-BiGRU-A algorithm, which combines BERT and BiGRU methodologies. Experimental results demonstrate a significant improvement in prediction accuracy, reaching 98.41% with the proposed approach. By leveraging advanced machine learning techniques and carefully managing data attributes, this study contributes to the development of more effective tools for financial market forecasting.