Stock Price Prediction Using LSTM with Attention-Based Multimodal Fusion Model

Stock price prediction is a highly challenging research area in finance, primarily due to the complexity of market data and the effective integration of multi-source information. This paper proposes a novel deep learning-based framework that employs a multi-level attention mechanism to achieve precise modeling of stock price movements. The framework can automatically identify critical time points, dynamically screen influential features, and integrate multi-modal information including market data, news sentiment, and market sentiment indicators, thereby enhancing prediction accuracy and stability. Experimental results demonstrate that this method delivers superior predictive performance across various market conditions, particularly showing strong adaptability and early warning capabilities during extreme market movements. This research provides reliable technical support for intelligent financial decision-making.