Our inspiration was in noticing how much popular news articles affect the price of stocks. We wanted to find a way to use that to improve our stock predictions
What it does
Uses natural language processing to create a "sentiment" feature to add to the stock price history. Then runs those through a Recurrent Neural Network, which is a specialize neural network which takes in time-series data. This enables our predictions to be intelligently based on the previous 200 days of stock valuation and sentiment rather than just a single day at a time.
How we built it
We used python for everything, the web front end was to be built in PHP using the Laravel framework, but we ran out of time.
Challenges we ran into
The biggest challenge was handling the large data sets and combining the stock data and the sentiment data in such a way as to be useful to the training algorithm.
Also, the time needed for training and natural language processing for so many data points means that if something goes wrong, we have to wait another half an hour to see if our fix worked.
The data we used was rather dated, as any stock market data set with live values required a paid subscription, so our free data set contained data up to March 23rd 2018.
What's next for Smart Sentiment Sensing Stock Service
We'd like to finish the front end, so that a user can go to our website and enter a stock ticker and receive a prediction for the next week's stock price.