With the modern day internet, it is very hard to decipher what is real and what is fake. This is not helped by one of the most powerful world constantly spouting "fake news", nor is it helped interference from foreign entities in the attempt to spread misinformation. Due to these issues, we were inspired to do something about it in the form of artificially intelligent article verification.
What it does
This program is either embedded into the chrome browser or accessible via python script to send news articles to be analyzed. The analysis is split into 4 different techniques across 2 layers. The first layer consists of sentiment analysis, factual verification and unsupervised word clustering. The second layer brings the outputs from the first layer and uses them as inputs to generate a full perspective on the accuracy of an article.
How we built it
Challenges we ran into
We ran into several across the stack. At the top level there were issues with jQuery and sending json rather non x-form-urlencoded requests, however that was eventually resolved. Another challenge was accurately building the training set in the very short amount of time allotted as well as tuning the system to be accurate.
Accomplishments that we're proud of
We are proud of the fact that it can distinguish some fake and real news from each other. We are also proud of the amount of work we pulled in off in the last 2 days!
What we learned
We learned that coding on pure caffeine makes one miss simple mistakes. To add to this, we also learned that GoogleCloud and AWS are similar, but their slight differences make the transition rather difficult. The other thing we learned is that jQuery can be extremely tricky when dealing with modern web development.
What's next for News Accuracy Analysis (Fake News Detector)
Optimize the code, train on more articles, add a web scraping component and make it into a chrome extension to be available to everyone.