Skip to content

Crypto Sentiment Analysis is a Python project using the Hugging Face transformers module and the Facebook RoBERTa model to predict sentiment in cryptocurrency news. The model processes recent news via a FastAPI endpoint, integrated with my Trading Platform

Notifications You must be signed in to change notification settings

JadoreThompson/CryptoSentimentAnalysis

Repository files navigation

Description

This project uses the transformers module from Hugging Face and the Facebook RoBERTa model to predict sentiment in cryptocurrency news.

Why BERT?

Unlike GPT and other well-known LLMs, BERT (Bidirectional Encoder Representations from Transformers) is bidirectional, meaning it can read text both forward and backward. This allows it to capture relationships between words across a passage, even if they appear far apart, making it especially effective for understanding context.

BERT was trained on Masked Language Modeling (MLM), which forces the model to predict missing words in text, enhancing its ability to generate strong contextual embeddings. Unlike static word embeddings, BERT dynamically generates embeddings based on context, so the word "free" might have different representations depending on usage.

This specific BERT model has only 300 million parameters, making it lightweight enough for my system while maintaining efficiency after removing the Next Sentence Prediction (NSP) task.

Prerequisites

  • Training: I recommend having a GPU, as training can be slow without one.
  • News API Key: The model is served through a FastAPI endpoint connected to my Trading Platform and uses the News API to fetch recent news data.

Installation

# Clone the repo
git clone <https://github.com/JadoreThompson/CryptoSentimentAnalysis.git>

# Install dependencies
pip install -r requirements.txt

Contact

If you're interested in collaborating or have any opportunities, feel free to contact me at [email protected] or connect with me on LinkedIn.

Evaluation

In the future I’m looking to build a list and phrases that indicate sentiment in an attempt to aid the model then combined with the CIMAWA model mentioned in the FBert paper. All in all this was a good experience for getting hands on with ML following my Property Valuation Project where I used Gemini to feed a chain of prompts gathering data from a Postgres DB and leveraging SERP to harbour more data. If you've cloned the repo fell free to contact me too I'd like to communicate with people who find my work interesting

About

Crypto Sentiment Analysis is a Python project using the Hugging Face transformers module and the Facebook RoBERTa model to predict sentiment in cryptocurrency news. The model processes recent news via a FastAPI endpoint, integrated with my Trading Platform

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages