BERT Model for Classification of Fake News using the Cloud Processing Capacity

This paper aims at conducting a predictive analysis on news articles in order to find if they are fake or real. After conducting an extensive research on the topic, various Machine Learning and Deep Learning models for the purpose of evaluating news articles were discovered. A new transfer learning model, Bi-directional Encoder Representation for Transformers (BERT), is tested using the Google Cloud GPU capacity for the purpose of detection. The first step in this direction will be to pre-process the data to clean out the garbage and missing values. After this, all the news articles collected will be tokenized, according to the BERT tokenizer. The tokenized corpus will be converted into tensors for the model to be trained. The data will be trained in batches with each batch having 32 articles. The final layer for training will consist of a five layered neural network. The model with the least validation loss will be tested for accuracy. The predictions for news articles will be made on this model. The paper will also explore the best cloud platform to host such a model and performance of the hosted model as well.