Enhancing Financial Sentiment Analysis with BERT and Data Augmentation for Market Impact Prediction

Abstract

This study explores the BERT-based financial text sentiment analysis method and evaluates the impact of different data augmentation strategies on sentiment classification performance. With the explosive growth of financial market information, how to extract effective sentiment information from text data such as news reports, market comments, and corporate announcements has become a key issue in market analysis and investment decision-making. The BERT-based deep learning model can capture complex contextual information and show significant advantages in financial sentiment analysis tasks. Experimental results show that BERT and its variants (such as FinBERT and RoBERTa) are superior to traditional NLP methods in financial text classification, while data augmentation strategies (such as synonym replacement, random deletion, and back translation) further improve the generalization ability of the model. Among them, the back translation method has the most obvious improvement on model performance, effectively improving classification accuracy, precision, and recall. In addition, this study combines LSTM for market impact prediction and verifies the correlation between financial text sentiment and market trends. Through time series modeling, sentiment analysis results can be used to predict market fluctuations and provide forward-looking decision support for investors. The results of this study show that the financial sentiment analysis method combining BERT and data enhancement can improve the accuracy of market sentiment monitoring and provide new technical support for smart investment advisors, quantitative trading, and financial risk management.