It is very important for people to follow the news because the news helps people keep up with the events around the world, and lets people be aware of what happen in our society. Our goal is to simplify this process. With our project you can watch all recent news all around the globe in one second.
What it does
Our project is all about collecting news from all around the web and preparing it for you.
We are using all the power machine learning offers to find similar packs of news and generate a custom headlines and summaries for them.
With the latest achievements of ML technology, like attention in RNN or Facebook's FastText vectorization we can do this quite well.
How we built it
We have used Reccurent Neural Network with Attention to generate headlines, TF_IDF feature extractions with K-means clustering to cluster articles. We have also tried more complex things but it didn't worked out.
Gensim models to tag news and news clusters as well as generate summaries for them.
Our web application is powered by React and Redux, built by Webpack.
Challenges we ran into
We had a couple of interesting problems with ML/AI part, for example attention thing wasn't working like it supposed to be.
Also, we've tried a lot of clusterization technics to find the best one for this job.
We had a lot of time with AWS platform, trying to increase limit on instances, chatting with tech support
Accomplishments that we're proud of
An awesome clusterization, good headline generation and really-really brilliant design.
What we learned
We all have learned a lot of things about ML technologies and AI tricks, explored pytorch library and databases things.
We have learned well that not always complex models are better than simpler models.
We've also learned how to launch an AWS project and maintain it
What's next for News Articles for Mature Folks
Train a better model, promote our site and find the ways to monetize this project :>