In this project, we developed a web application called “verifyit” which checks the authenticity of tweets using the BERT model and also allows us to detect the super-spreader through influence score. The application deals with two objectives: A. To detect fake news: Implement a fake news detection model using Bert and compare the performance with covid19 fake news dataset(Codalab) and Generalised fake news dataset(FakeNewsNet). B. To trace out the super-spreader of this fake news: In order to consider a user to be a super-spreader, the tweets are analysed and evaluated based on their impact score, and if half of their posts are fake, their influence score will be increased and they are considered to be a super-spreader.
Dataset
Model Evaluation
Libraries required to run the program:
please install using the following commands:
!pip install pandas
!pip install tf-models-nightly
!pip install tensorflow-text-nightly
!pip install tf-models-official==2.5.0
!pip install tensorflow-text==2.5.0
!pip install tweepy
!pip install tensorflow_hub
!pip install trafilatura
If you wish run this program on google colab,Changing the runtime type to GPU significantly improves the training time if you wish to train the model again.
Please change the file locations accordingly as per file location saved in your colab directory.
To access tweepy connection, please generate a twitter elevated developer access through the documentation provided below
https://developer.twitter.com/en/docs/twitter-api/getting-started/getting-access-to-the-twitter-api
Once access has been generated, replace the comment for the token variables provided in the notebook.
To access ngrok connection, please generate an auth token through the link provided below: https://ngrok.com/docs/secure-tunnels#tunnel-authtokens
Once token has been generated, replace the comment for the token variable provided in the Fake_News_Main notebook.