Social Media Sentiment Analysis Using a Fine-tuned Large Language Model

• This project is an initial preparation phase for social media sentiment analysis using a fine-tuned large language model. There are certain shortcomings, which will be addressed and improved in the future.

Project Information:

• Author (in no particular order): Yiwei Liang; Jiaolun Zhou.
• Instructor: Prof.Luyao Zhang
• Project Summary: At this stage, the project uses the Llama2 , LoRA method and the Alpaca LoRA model for the study, based on user comments collected from online forums to train a model capable of making an analysis of cryptal price data based on daily forum data. The model is not able to analyze the cryptal price quantitatively, but only able to make "positive", "negative" and "neutral" judgments on the cryptal price trend. In the future, larger orders of magnitude of data and Llama30b can be used to correct this limitation.

Upon completion of this project, it is expected to be put into practical production in the following areas:

• Real-time market insights: By analyzing forum data, our model can provide implementation insights into the virtual currency market, including investor sentiment, the impact of news events, and possible price fluctuations.
• Risk Management and Decision Support: Financial institutions, investors and traders can utilize our model to assess investment risk, specify trading strategies and make more accurate investment decisions.
• Early Warning System: Our model can be developed into an early warning system that identifies market anomalies and potential risk events in advance, helping market participants, policy makers, and academics to adjust their strategies in a timely manner.
• Educational and Research Tools: Our model can be used as an educational and research tool to help students, academics and researchers better understand the dynamics and influences of the virtual currency market.
• Industry standard and innovation: Our model can become the new standard for virtual currency price prediction and drive technological innovation and application in related fields.
• Extension to other financial markets: Our model can be extended to other financial markets. Successful virtual currency price prediction methods can help make accurate predictions for stocks, bonds, commodities, and other areas, providing valuable assistance to the broader financial sector. Overall, our project has great potential for the industrial sector, and at the same time, our project is innovative, and at this stage, we are not able to search Google Scholar for studies related to putting the Alpaca LoRA model into financial analysis and forecasting.

Table Of Contents:

1. Self Introduction
2. Research Questions
   • Research questions
   • Significance
   • Study 1: "Using GPT-3 for Stock Market Prediction"
   • Study 2: "Cryptocurrency price prediction using traditional statistical and machine-learning techniques: A survey"
   • Study 3: "Forecasting the price of Bitcoin using deep learning"
3. Methodology
   • Introduction to Methodology
   • Llama2 model
   • Transformer architecturel
   • LoRA method
   • Alpaca-LoRA model
   • Data process
   • Data Analysis
   • Limitation
   • Future research
4. Application Scenario
5. Data
6. Code
7. Result

More About the Authors

Yiwei Liang

• 

Self-introduction:

I am Yiwei Liang.

Jiaolun Zhou

• 

Self-introduction:

I am Jiaolun Zhou from Computer Science at Duke Kunshan University, working on IoT and machine learning, and I am honored to be able to participate in this research and contribute to virtual currency price prediction and the future development of the financial market.