CryptoVibes

We are building a web application that extracts transcripts from trending crypto YouTube videos and uses ChatGPT Prompt Engineering to compile a visual and insightful trading recommendation system.

Comment

Inspiration

We are convinced that online social media post sentiments can successfully predict the close movement of the stock market. We are enthusiastic traders and have experienced a more-than-50% correlation between social media (twitter, YouTube, Reddit, etc) post sentiments and the next-day movement of the stock market. This is highly intuitive since more people buy a certain stock/crypto, the higher the price increase.

What it does

The backend scrapes a good amount of trending crypto influencer videos between the length of 30 seconds to 30 minutes, extracts the transcript, and does sentiment analysis and investment recommendation summaries for the ten most mentioned cryptocurrencies for each video with ChatGPT Prompt Engineering. After this, a weighted score of sentiments and crypto relevance is calculated and these scores and recommendations are stored in a JSON file and Mongo DataBase before they are visualized as a HeatMap of the top ten most relevant cryptocurrencies on our website, localhost:3000. A summary of the project and a "leaderboard" of the most trending cryptocurrencies with sentiment scores (0 to 10) and stockmarket predictions can also be found on our website.

How we built it

We have started building the backend (Python YouTube video scraping and transcript extraction system, ChatGPT Prompt Engineering which outputs a JSON file stored in a Mongo database) and the React Frontend at the same time. After the data storage and manipulation pipelines were ready, we finalized the information flow by populating the front-end data visualization tools.

Challenges we ran into

Some videos were in a different language than English and some of them were too long to be used for ChatGPT Prompt Engineering, hence we had to filter out these videos.

Accomplishments that we're proud of

Yesterday the data pipeline highlighted WEN and ShibaInyu which skyrocketed by 50% by today. This is an excellent proof of concept that this tool has a real-life potential to predict the movement of the stock market and aid traders at multiple experience levels in their investment journey. This tool is also highly appealing to look at, user-friendly and gamified.

What we learned

We learned that Mongo DB might be an overkill with a system of this magnitude. We learned to utilize data manipulation and front-end tools.

What's next for CryptoVibes

We have to implement a more sensitive sentiment analysis system and a way to measure success by introducing KPIs and accuracy measurement systems. We have to validate the accuracy of actual stock prices and check it against our predictions. Extracting and visualizing further relevant data from the YouTube scripts, implementing a subscription system, buying our web domain, making the data pipeline more robust, and implementing a daily auto-refresh system to keep the information up-to-date are also possible value-adding activities for the future. Improving the relevant YouTube video selection system, and populating the website with text-based investment recommendations by trending cryptocurrencies and in general are also further nice-to-haves.

Built With

Share this project:

Updates