Enhancing Stock Price Prediction Through Fine-Tuned CardiffNLP Twitter-RoBERTa: A Web Application Approach Using Financial News Articles

This paper introduces a novel method for predicting stock price movements by fine-tuning the CardiffNLP Twitter-RoBERTa model-a pretrained language model specialized in social media sentiment analysis-for financial news summaries. Recognizing the significant impact of news sentiment on markets, we aim to enhance an existing web application by providing rapid, real-time sentiment assessments of company news, offering valuable insights to investors. Our key innovation lies in adapting a model designed for informal, sentiment-rich social media content to the formal context of financial news. Despite stylistic differences, financial news summaries share structural traits with social media posts, such as brevity and high sentiment density, facilitating this effective adaptation. This approach introduces a unique perspective to stock price prediction by applying advanced NLP techniques from social media analysis to the financial domain. Our fine-tuned model outperforms all baseline models, including BERTweet, FinBERT, and FinancialBERT. The fine-tuned model was integrated into a real-time web application built with Angular and Node.js, enabling instantaneous sentiment evaluation through a web-based API. This system empowers users with timely, actionable intelligence for stock movement analysis, facilitating informed investment decisions. This work contributes to financial sentiment analysis and highlights the potential of cross-domain model adaptation for enhancing investment decision-making.