Stock Price Movement Prediction Using Sentiment Analysis and CandleStick Chart Representation

Area of Science:

• Computational finance
• Machine learning applications
• Financial market analysis

Background:

• Stock price prediction is complex due to numerous influencing factors.
• Existing machine learning models struggle with single data sources for accurate stock movement prediction.
• Integrating diverse data streams is crucial for enhancing predictive capabilities.

Purpose of the Study:

• To propose a novel multichannel collaborative network for stock trend prediction.
• To integrate social media sentiment and candlestick chart data for improved forecasting.
• To evaluate the model's performance across different time horizons and stocks.

Main Methods:

• Utilized Natural Language Toolkit and sentiment analysis on Twitter data for social media sentiment extraction.
• Transformed historical time series data into candlestick charts for pattern recognition.
• Developed a two-branch network with 1D CNN for sentiment classification and 2D CNN for candlestick chart image classification.
• Integrated features from both branches for stock price movement prediction.

Main Results:

• The multichannel collaborative network demonstrated superior performance compared to single-network models.
• Achieved a maximum accuracy of 75.38% for Apple stock prediction.
• Observed improved prediction accuracy for longer time periods (10 days) versus shorter periods.

Conclusions:

• The proposed collaborative network effectively combines social media sentiment and visual stock patterns for accurate price movement prediction.
• The model offers a promising approach for enhancing stock market forecasting by leveraging multimodal data.
• Future research could explore further integration of diverse data sources and advanced deep learning architectures.