Multimodal deep learning for stock market prediction
- Publication Type:
- Thesis
- Issue Date:
- 2024
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Stock market movements are influenced by public and private information that includes historical market data, news articles, company reports, and social media discussions. Analyzing these vast sources of data can give market participants an edge to make profits. However, the majority of the studies in the literature are based on traditional approaches that come short in analyzing unstructured, vast multimodal data. Also, most existing works focus on singular data sources, which end up missing huge amounts of information from other data sources. Recently, while multimodal forecast models have been proposed, they mostly focus on either modality-specific or joint information in the modalities. Although there is a great opportunity for the use of multiple data sources for market analysis, challenges exist in efficiently modelling these modalities together. Another challenge lies in the multi-modal representation of these multiple modalities and capturing cross-modal information from the input data. To address the challenges of multi-modal learning in stock market prediction, we propose several innovative deep learning based learning models that can utilize both modality-specific and joint information across modalities. The main contributions of these models lie in their effective utilization of various data sources for price movement prediction, with advanced multi-modal representations and data feature fusion techniques. We propose novel models to predict both closing prices (regression models) and directional movements (classification models). We evaluate the performances using various metrics (accuracy, MCC for classification; MAE, MAPE, MDAPE, RMSE for regression) to evaluate the performances of our models and compare them against the state of the art analysis models. In the experiments, our proposed novel multimodal models outperform the baseline models and improve the prediction performances. Our experiments with real-world datasets show the effectiveness of the proposed multi-modal deep learning designs in stock market analysis and forecasting.
Please use this identifier to cite or link to this item: