A1196
Title: GNN based social medium analysis in stock prediction
Authors: Pengju Zhang - University of Bristol (United Kingdom)
Richard Harris - University of Bristol (United Kingdom)
Jin Zheng - University of Bristol (United Kingdom) [presenting]
Abstract: The purpose is to introduce a novel approach that utilises graph neural networks (GNNs) for sentiment analysis to enhance stock market predictions. By integrating social media sentiment analysis with traditional financial indicators, a comprehensive model that accurately captures market sentiment dynamics is developed. The methodology includes data collection from social media platforms, sentiment extraction using GNN, and prediction using advanced time-series models such as LSTM, CNN and Transformer. The model is evaluated on several stock datasets, and improvements are demonstrated in prediction accuracy and trading performance compared to traditional models. These results underscore the value of merging sentiment analysis with time-series techniques for financial market prediction. Furthermore, the use of the prediction model to improve trading strategy performance is discussed by comparing different trading strategies based on the daily return and Sharpe ratio. The contribution to the field is by providing insights into the role of sentiment in financial markets and by advancing the capabilities of predictive models through the integration of GNN-based sentiment analysis.
