Harnessing the power of social media sentiment analysis, we can take user submissions across cryptocurrency subreddits and create a model that can predict cryptocurrency prices.
Inspiration: I myself am really into cryptocurrencies and their potential use. I hypothesized that we could predict crypto prices with social media because 3 reasons:
Cryptocurrency market is adopted more by retail investors rather than institutional ones. Allowing for less BTS market manipulation.
Crypto market is more unregulated and open 24/7, allowing for fast trading and at any time.
Reddit holds niche but effective subreddits that talk about these cryptocurrencies.
What it does: It predicts the average price of BTC/ETH from a set list of features scraped from subreddit posts, comments, and sentiment scores. This set of features is then applied to a Neural Network Regression model to build a model that accurately predicts daily average price((24hr low + 24 hr high)/2). Based on the model, it would show if my hypothesis is correct or not based on this approach.
How we built it: I used jupyter notebook to create python files that can:
Scrape all necessary data from Reddit with PRAW
Convert datasets into CSV files with Pandas
Run CSV files through VADER sentiment analysis, attach scores
Build neural network model with PyTorch
Challenges we ran into: The biggest problem I ran into was the lack of time. The dataset scraping in itself took ~4 hours to do. If I had more efficient code or a faster processor, I could decrease that time which would allow me to work on the rest of the project.
Accomplishments that we're proud of: Scraping 1 year's worth of comments from top posts in Bitcoin and Ethereum Subreddits. Running some of the CSV files through VADER sentiment analysis to find scores of comments. Designing project from end-to-end and determining which type of ML model is the best use.
What we learned: I learned how to scrape Reddit and work with Pandas. I also learned a lot about Sentiment Analysis(Lexicon vs. Semantic), picking the right ML model based on our project's scope. It would be best if you definitely accounted for the time needed to scrape, compute, and train.
What's next for Crypto Market Predictor: I want to run the neural network and run the test set to see what the performance of the model is. I would also like to include and scrape data for other popular cryptocurrencies like XRP, ADA, XLM, or even DOGE!
Log in or sign up for Devpost to join the conversation.