-
Welcome to Dashboard! Enter a stock symbol in the search bar...
-
We entered SNAP, looks like Snap, Inc. is doing well in the news today!
-
We also checked on AMD, looks like the news articles aren't very positive about AMD today...
The stock market is complicated, we make it simple. All you need to know: positive or negative?
Inspiration
One of the (arguably the most) important factor is day to day, hour to hour, stock decision comes from the performance of the company. You can spend time skimming multiple news networks and articles to get an idea on how your company is doing, but can we make it faster and simpler?
What it does
When you navigate to our webpage, you see a search bar, where you can enter a valid stock symbol (SNAP, AMD, APPL, etc). Hit enter, then after a few moments, the page updates with an image of a plus or a minus sign. Seems incredibly simple, right? What if I told you that during those few moments it takes for the webpage to load after you search, our app is doing complex NLP to give you an overall positive or negative sentiment about the company you searched for?
Pretty cool right?
How we built it
Web Application
A simple and efficient full stack application, powered by EJS and node.js.
Classification Algorithm
Now this is the brain of the app. We trained a Naive-Bayes classifier on a large data set consisting of article headlines, to detect positive or negative sentiment. After training the classifier, it was able to take in a sample headline, and classify it as positive or negative, with a high accuracy rating. So now, with this constructed classifier, when you enter a stock symbol, our application scrapes for all news articles relating to the company from the past hour, runs their headlines through the classifier, and provides you with a weighted sentiment on the overall state of the company: + or -. Our app provides you an NLP-powered, real time metric on the state of any public corporation.
Challenges we ran into
One of the biggest challenges in data science is that your final result is only as accurate as your training data. Due to the limitations of a 24-hour hackathon and a small 2-person team, building a high quality, manually-tagged data set for our classifier to train on proved to be a challenge. To solve this, we resorted to a smaller data set, with records sourced from public data sites. Going forward, with more time and manpower, we would like to mine a higher-quality data set, to improve the accuracy of our classifier.
Accomplishments that we're proud of
A working full stack application! This project blends both of our skills uniquely and perfectly, and we were able to produce a finished final product that utilizes a variety of tools and technologies.
A working data science project! This application is a full machine learning project from start to finish, from data mining, data preprocessing, model training, and application of the model.
What's next for Dashboard
As we mentioned above, we want to tackle our data mining problem. Additionally, we want to improve on our frontend and add a few more features (not too many though, keep it simple of course).
Log in or sign up for Devpost to join the conversation.