What it does
Trendcast takes sentiment analysis data of tech articles from major tech news sources along with Google Trends data to generate an overall rating over time based on popularity and overall sentiment. Machine learning is then used to predict future trends by training past data on Long Short Term Memory networks.
How we built it
The front end was built using React and is hosted on Firebase. The back end used Python to scrape news articles from archives using Beautiful Soup, analyzed sentiment with NLTK, a natural language processing toolkit. The time series predictions were done using Keras and Tensorflow models with recurrent neural networks.
It was difficult and frustrating to scrape so much data over really slow WiFi. It was also our first time trying out machine learning libraries and we were all unfamiliar with data science. Aside from some major technical challenges, the volatility of cryptocurrency prices combined with the newness of the technology made it difficult to predict future trends with a high degree of confidence.
Accomplishments that we're proud of
We are very proud of actually implementing the idea, putting the front end together with an elegant UI, hooking everything up to Firebase, scraping and processing a ton of data and getting the prediction model to work.
What we learned
Through this project, we learned a lot about data management and analysis, machine learning libraries such as Tensorflow, new trending technologies and full stack web development.
We want to add more trending technology categories and a live news feed for current topics. This could be done with a real-time news aggregating model that can collect and analyze data from major news sources and APIs automatically.