Currently, I am leading a thesis paper on applying LLMs to market sentiment predictions on market news. In contrast to previous research, this paper will take a deeper dive into the technical aspects of LLMs in hopes of improving accuracy against even the highest-performing general-purpose model such as ChatGPT4. To do this, I created a new dataset based on Dow Jones News Wires and Wharton Research Data Service with 120,000+ entries. Then finetuned existing LLMs such as Bert, Mistral, and Llama using Huggingface Transformers, DeepSpeed, and LoRA for sentiment analysis on the newly created dataset mentioned above.
Copyright © 2024 WYETT