Over the past few months, the news has been more prominent in our lives than ever. In addition to this rush of information, misinformation seems to be even more prominent, with the phrase "fake news" becoming part of our everyday vocabulary. It is for this reason that a method is needed for people to better keep up with current events in this uncertain time.
What it does
It is a machine learning powered aggregator of news stories. It aims to allow people to quickly develop a basic and accurate understanding of current events. The website allows users to browse through summarized versions of long articles, explore topics or cross reference documents through automatically labeled documents, or analyze the trustworthyness of information through detections of biased language. In addition to allowing users to browse through articles with these features, all of these technologies are combined into a single, user friendly web app where the user can simply enter how much time they have, and receive a "briefing" containing articles that contain the widest variety of topics and are as unbiased as possible.
How I built it
The webserver is hosted using Flask. The automatic topic labeling is done through an unsupervised machine learning algorithm called Latent Dirichlet Allocation, using the gensim library. The model was trained of 4000 scraped news articles about the coronavirus. The promotional language model is a deep learning model called an LSTM, the full architecture is on https://github.com/AlexWan0/Your-Covid-Briefing.
Challenges I ran into
Web design is something that I'm still new at, creating a responsive design took some time.
Accomplishments that I'm proud of
The promotional language detector worked surprisingly well, it ended up being more accurate than I expected.
What I learned
I have never used Latent Dirichlet Allocation before, so creating that was a great experience. Through this, I also I strengthened my skills in other areas like web design.
What's next for Your Covid Briefing
I want to focus on improving the efficiency and accuracy of the promotional text model, there's still some room for improvement. Some aspects of the web design still need polishing as well.