Enhancing Financial Forecasting through Transformer Models Using Social Media and News Insights

Gurijala, Saiharsha (2024) Enhancing Financial Forecasting through Transformer Models Using Social Media and News Insights. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (793kB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

Predicting the stock market correctly is challenging as it is dependent on corporate fundamentals, macroeconomic factors, and market sentiment. This study presents a new model combining Temporal Fusion Transformers (TFT) and sentiment analysis for stock price prediction. Inference is conducted based on the historical data observed for many structured features along with the time-sequenced sentiment obtained from discussions on Reddit Financial Sub-reddit and financial news, which is fetched to perform Finbert, a specialized NLP for financial data points. The prediction between the two Long Short-Term Memory (LSTM) models used here shows that the TFT model has the superpower to radically higher Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and R² Score than a traditional LSTM model. Furthermore, the model's interpretability capabilities, including attention layers and feature importance assessments, enhance understanding of the underlying drivers of stock price forecasts. These results underscore the importance of sentiment-driven features in terms of prediction accuracy, which renders their practical significance apparent. This research presents a scalable, interpretable approach to financial forecasting using advanced transformer-based architectures and unconventional data. This framework is a critical step toward connecting academic research with practical applications, offering tools for algorithmic trading and decision-making. This means their work is not only important for upcoming advances in AI in finance, but also for the use of AI more generally, as we try to find ways to use systems that are accurate but also interpretable to help us guide the muddle of complex modern finance.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Anant, Aaloka UNSPECIFIED
Uncontrolled Keywords:	Temporal Fusion Transformers (TFT); Long Short-Term Memory (LSTM); Financial Forecasting; FinBERT; Natural Language Processing
Subjects:	H Social Sciences > HG Finance Q Science > QA Mathematics > Electronic computers. Computer science T Technology > T Technology (General) > Information Technology > Electronic computers. Computer science P Language and Literature > P Philology. Linguistics > Computational linguistics. Natural language processing Z Bibliography. Library Science. Information Resources > ZA Information resources > ZA4150 Computer Network Resources > The Internet > World Wide Web > Websites > Online social networks T Technology > TK Electrical engineering. Electronics. Nuclear engineering > Telecommunications > The Internet > World Wide Web > Websites > Online social networks H Social Sciences > HG Finance > Investment > Stock Exchange
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	02 Sep 2025 12:22
Last Modified:	02 Sep 2025 12:22
URI:	https://norma.ncirl.ie/id/eprint/8707

Actions (login required)

View Item