The problem Neural News solves

Media should serve as a source of correct information, but they are heavily supportive towards one ideology and cause. Hence, the news is published in a biased way and we cannot neglect it, because it affects the mindset of billions of people.

So, we aim to build a web-based community where people share articles about their own locality or field and our ML model predicts whether it is biassed towards one side or not and gives a score showing its likelihood to be unbiased.

The community members can review the news and upvote or downvote the article and reviews to ensure that the article we received is correct.

To tackle the problem of spamming, fake reviews, or any argument we set certain restrictions such as the ones on StackOverflow. Such that only people with more than 200 reputations can post articles, and only the ones with more than 50 reputations can flag or upvote/ downvote an article or review or can comment on the review. Similarly, the review must follow answering guidelines.

Hence, we look forward to completing every bit of this project and to deploy it with some actual users. We also thought of various incentives we can add for our top contributors like youtube in order to make it possible.

(Sorry for the last minute hustle, but we are deploying it on Heroku and its link will be available on Github's readme.)

Challenges we ran into

Getting a research article already done on this stuff was tough as there are many models on Fake News classifier but not even one on true news loaded with a bias towards on side.

We had trouble searching for the website of news articles which will be easy to scrap using BeautifulSoup. We found India Today and Reuters to be good enough for it.

Accuracy of the Model was quite good on test data but when we added real-time data the prediction was very inconsistent. Hence we were introduced to NLP by mentor Shridhar and we learned about stopwords and it improved the consistency by a lot.

We also struggled with model selection and tried a lot of them but obtained the best results from LSTM and logistic regression.