Current growth trends and popularity of cryptocurrency and blockchain technologies.
What it does
A collection of data, as a proof of concept that would allow us to correlate twitter user sentiment with bitcoin (and other cryptocurrency) price. Currently, we've collected a mass amount of data related to bitcoin, filtering it with good and bad keywords in order to train a neural net.
How we built it
First, we compiled a list of positive and negative words/jargon in cryptocurrency. Then, we obtained 100 tweets per week for each category for the last 4 years. After that, we gathered 100 tweets per day for all of bitcoin-related tweets over the last 4 years. Our search query was something like "bitcoin -ethereum -litecoin". Meanwhile, one of our teammates prepared the model for sentiment analysis using the 'supervised' training data. We generated sentiment pair values for each tweet, and started building a website to interface with them.
Challenges we ran into
Since we didn't have access to AWS for the GPU instance, we spent almost the entire hackathon trying to generate our deep learning data. Additionally, learning how to scrape mass amounts of data and manage it gracefully required a lot of backtracking.
Accomplishments that we're proud of
We have a primitive neural network that can be improved with more test data and computing power. We also were able to make a lot of progress collecting, cleaning, and utilizing the data.
What we learned
We learned a lot about Twitter's API, python data handling, and various APIs
What's next for Bitcoin Twitter Mining
This project can easily be used as a reference or checkpoint for future endeavors in both sentiment analysis and with deep learning. Since the cryptocurrency market is so volatile and sensitive, the emotions we would be able to capture through tweets would supplement well with regular cryptocurrency market analysis.