This dissertation was written as a part of the MSc in Data Science at the International Hellenic University.
It presents a thorough investigation into the use of a Double Deep Q-Network reinforcement learning
model, which leverages Technical Indicators and Sentiment Analysis extracted from the social media
platform StockTwits, to forecast stock market trends. The case study in this thesis is NVIDIA stock,
which demonstrated non-stationary similar behavior with a clear upward trend from January 2, 2020, to
September 21, 2023.
Financial data for the analysis were sourced from Yahoo Finance, with technical indicators computed via
the yfinance Python library. For the sentiment analysis, the 'Twitter-roBERTa-base for Sentiment Analysis'
model was employed, a tool that has been rigorously trained on roughly 58 million tweets and fine-tuned
for sentiment classification using the TweetEval framework. Daily sentiment data were aggregated from
StockTwits, capturing the market's pulse in response to news events, price changes, and general
sentiment. The outcomes of this study are presented in the relevant chapter of this study and underscore
the promising future of reinforcement learning models in stock market prediction, especially those that
integrate sentiment analysis, indicating a transformative step forward for algorithmic trading approaches.
Collections
Show Collections