Targeted Financial-Oriented Social Media Sentiment Measurement

This research article presents a natural language processing (NLP) model designed to measure financial-oriented sentiment directed at specific firms within social media texts. The study begins by creating a human-annotated dataset focused on targeted financial sentiment, which is crucial for accurately analyzing social media posts that often contain mixed sentiments about various entities. The authors propose a prompt-based model architecture that demonstrates superior performance on benchmark datasets for targeted sentiment analysis. The model is fine-tuned using the annotated dataset to enhance its accuracy in measuring sentiment towards 24 meme stocks and 30 Dow Jones constituent stocks from a dataset of 23 million social media posts. Results indicate that the sentiment measured by the model correlates positively with price returns and negatively with price volatility, particularly for meme stocks. The study also compares the model's performance against other financial sentiment measurements, showing its economic superiority in predicting trading strategy returns.