The proliferation of fake news on social media and internet is deceiving people to an extent which certainly needs to stopped.
The main objective of this project is to generate a model that can discriminate between fake and true news articles when it is trained with a certain dataset.
-
The Dataset is acquired through Kaggle which had two different csv files namely true.csv and fake.csv.
-
Merged both of the dataset in random fashion to create a complete whole dataset by adding a new column called label which had ‘fake’ value as 1 and ‘true’ value as 0.
-
Pre-processing the dataset using Bert Tokenizer for splitting the news sentences into meaningful smaller pieces called tokens.
-
Defining BERT Model and it's architecture by freezing some layers and fine tuning it from scratch using Deep Learning.
-
Defining the optimizer , Train and Evaluation function.
-
Designing a simple easy to use interface using Gradio Library and Predicting and calculating the performance using compute metric function.