Fake Content detection using RNN

Inspiration

This project was inspired by a research paper that I came across recently which uses a combination of naive Bayes, SVM, and semantic analysis to predict fake news.

What it does

My model takes the news as input and out says if the news is real/fake.

How we built it

Dataset used I used the Kaggle dataset :link for training.

Preprocessing For text pre-processing, I used functions from Tensorflow. This includes removing unwanted words in the news like is, that, are, etc, and removing special characters like $, %, ! which doesn't make sense to the news. Then I did tokenization, stemming, etc that needs to be done before using any NLP model.

Selecting the model I collected few neural network algorithms like CNN, RNN, and ML algorithms like logistic regression, Multivariate Bayes classification (also tuned hyper-parameter), XGboost, random forest, and decision tree for the prediction. I tested it on the Kaggle dataset found that RNN gave a maximum accuracy of over 0.99

Challenges we ran into

It was difficult to first search for the best dataset. Since I am new to DL, it was difficult to understand neural networks and the way it works. Everything else was a cakewalk.

Accomplishments that we're proud of

I aim to develop this model further to help in national security.

What we learned

I learned more about neural networks like CNN, RNN and using a combination of ML algo like Naive Bayes and SVM.

What's next for Fake Content detection using RNN

I wish to enlarge my domain of fake news classification for social media posts and aim to achieve further accuracy

Built With

• matplotlib
• numpy
• pandas
• python
• seaborn
• skikit
• tensorflow