• disinterest in some of the millennials
  • Less knowledge of applications where to find such news
  • lack of good, unbiased and unfiltered content to read about current affairs
  • lack of content which gives in depth knowledge of the truth

What it does

  • It gives an personalised experience for the user by giving him recommendations based on his previous swipes
  • It makes an user aware about current important political conditions by making sure the content is recommended at --- the correct time and between the correct news such that the user is not gets interested in reading it
  • taking into account the value of time in a millennials life we summarise the article by highlighting the important parts of the article but at the same time displaying the entire content such that the deepness of article is not tampered
  • We value the time of our customers thus we give an total reading time based on the numbers of words in the highlights

How we built it

  • Database - We are using FireBase as our database

  • Interface - We have used vanilla javascript and animate.css to create the user interface

  • Recommendation System - We created a training dataset by observing usual user browsing patterns over news category. Then we created a LSTM ( Long Short Term Neural Network) to understand a users pattern and create a recommendation system. The platform we used here is Python.

  • Text Summariser/ Highlighter - We used TextRank algorithm - a graph based ranking algorithm used for text summarisation.

Challenges we ran into

  • We were given only URL of articles in the data thus we had to use web crawling for extracting articles and saving the entire HTML along with styling of the article in the database
  • We had very less training data for training our LSTM - Recommender model
  • The Python program for Text Highlighting returned lines from the article as an output, it was really difficult to find the exact lines in the HTML file of the article

Accomplishments that we're proud of

  • We created an real time application which was fetching company data from the database and using ML algorithms to give real time predictions and summarisations
  • Our Recommendation system gave us a really good prediction based on the minimal data which we had
  • Our Text Highlighter was integrated with the python code and the front-end and was working in real time

What we learned

  • We can achieve everything via teamwork and proper planning
  • LSTM's is a better model over Collaborative filtering when it comes to sequential data
  • We tried getting activation weights of an LSTM model to get the text summary but the graph based TextRank algorithm gave us more accuracy

What's next for Mirror

  • The application is real time and ready to use
  • We need to optimise the time for the recommendation
  • We need to retrain our recommendation engine on a longer training data
  • Current we are fetching 10 feeds from the database at a time (based on recommended categories) we plan to fetch it after each feed

Built With

Share this project: