Inspiration
First, we were just generally brain storming ideas. Then because of a general frustration with being able to tell what to invest in, we decided to solve that problem by seeing what people are investing in now.
What it does
TradeTone scrapes Yahoo Finance and Google News for articles about the given Tick input and uses the neural network FinBert, which is trained specifically to analyze whether financial information is being discussed positively or negatively, returning a vector of how positive or negative it is. Then we present the processed data back to the client with multiple metrics.
How I built it
We first held a vote on the multiple ideas we generated during the brainstorming step. Once it was decided that we were developing TradeTone we divided up the work, with one of us working on the machine learning and the other 2 on scraping the different websites. One was tasked with gathering the links which through much trial and error were gotten with yfiance an opensource tool to retrieve data in the stock and get access to its relevant . The using separate methods like beautifulsoup and request I retrieved the links from the google search results using the tick as an input in the search. We then inputted the list of links into the other set of code which used beautifulsoup to get all of the data in the 'p' tag in HTML. We progressively refined this to make it more accurate. We also went through multiple machine learning systems because none could get high accuracy within our test cases. Once we started using FinBert we started to see high levels of accuracy. Finally, we incorporated into its own website so that it is easy to interact and call the function with and can provide the data in a clear readable format for the user.
Challenges I ran into
We really struggled to make github work it was a big challenge. We also had to try multiple attempts to successfully use the machine learning algorithms and to accurately scrape the web and news sites. Our finally challenge was runtime as there was much we could do about it except lower the amount of inputs and speed up the code where we could
Accomplishments that I'm proud of
Very proud of our perseverance throughout this challenge we all didn't sleep so were working essentially 24 hours straight and were working with things we had never used and yet we were still able to make a really cool functioning design.
What I learned
We learned a wide range of things, what are the best models for analyzing financial text, how to scrape the web effectively, how to do the back end of a website, and how to make complicated data appear presentable
What's next for TradeTone
We hope to expand the number of websites that we search through so that we can have broader sets of data in order to be more accurate in our analysis. We would also like to start storing the data over time so we can begin to see larger trends in opinions toward stocks and provide that information to the user.
Log in or sign up for Devpost to join the conversation.