The study presents a novel approach for stock price prediction by integrating financial news articles into a deep neural network model called DP-LSTM. The model incorporates sentiment analysis of news articles using the VADER model and applies differential privacy to enhance prediction accuracy and robustness. The sentiment-ARMA model is introduced to combine stock price data and sentiment analysis results for improved prediction performance.
The paper discusses the challenges of stock prediction and the importance of incorporating various factors such as financial news sentiment. It highlights the need for automated sentiment analysis due to the vast amount of unstructured text data and the difficulty in manually analyzing it. The VADER model is utilized for sentiment analysis, providing polarity scores for news articles.
To address potential bias from subjective news reports, the study introduces the concept of differential privacy, which adds random noise to the prediction process to protect privacy and improve model robustness. The proposed DP-LSTM neural network architecture combines LSTM, sentiment analysis scores, and differential privacy mechanisms to predict stock prices accurately.
Experimental results on S&P 500 stocks demonstrate the effectiveness of the DP-LSTM model in improving prediction accuracy and robustness compared to traditional LSTM models. The DP-LSTM approach achieves a significant improvement in mean prediction accuracy and reduction in mean squared error for stock price prediction.
Overall, the study contributes to the field of stock price prediction by proposing an innovative approach that leverages deep learning, sentiment analysis, and differential privacy techniques to enhance prediction accuracy and mitigate biases from subjective news reports.
Read the whole study HERE