Although the J.P. Morgan Challenge certainly influenced our decision to build this application, we also saw firsthand the devastating effects of Hurricane Harvey around our university. Following this, we became determined to find a more efficient way of allocating emergency services to the areas most in need. With Big Data and Machine Learning on the rise, we discovered that a combination of these technologies were the perfect way to deliver our vision.
What it does
Audros begins by prompting a user for a location - they can select a state, then a county, and then a map appears of their selected location. On this map is a visual representation of the areas most affected by a natural disaster - points are plotted across the map based on information streamed from Twitter in real-time. Anytime that a tweet is posted which is determined not to be spam by our Machine Learning algorithm, Audros plots a point on the map using geolocation data obtained from the tweet.
How we built it
We used the Shiny framework of R to build the frontend, and used Python to create the services used in the backend. To implement Machine Learning, we used the sklearn library and determined that the SVM Classifier would provide the best results for our dataset.
Challenges we ran into
Building Audros was no walk in the park. But we knew that perseverance was a quality necessary for success on our project. From inexperience in Machine Learning and backend technologies to issues with the Twitter API, our program was no stranger to technical difficulties and bugs. However, we solved our problems every time by employing the help of our trusty friends Google, StackOverflow, and most importantly: collaboration.
Accomplishments that we're proud of
Walking into HackRice on Friday, we never would have thought that we'd build a fully-functional map system capable of classification through an SVM Machine Learning model, complete with twitter metadata and geolocations. The biggest thing that we'll walk away from this Hackathon with is definitely experience and pride - pride in our hard work and end product, which has the potential to help so many people across the United States.
What we learned
We learned a lot about ourselves this weekend - areas of Computer Science in which we're strong in, as well as areas where we have room to grow and develop. We'll spend the next few weeks diving into the documentation of Python, R, and Tweepy to build our knowledge of these areas, as well as immerse ourselves in articles on Machine Learning and Data Science to bolster our technical capabilities for the next Hackathon we attend. In all, each of us realized that there is so much to explore across all fields of Computer Science, from Machine Learning to frontend UX - and all of it is within our reach as Rice CS Students.
What's next for Audros
The next steps for Audros are twofold: first is an improvement in our Machine Learning algorithm, most likely a move from our current SVM Classification model to an Unsupervised model, using technologies such as Google's word2vec. The other step for our project is to gather more data which we can use to train our model. The more data that our model receives for training, the better it will be for classifying tweets as spam or relevant.