Fake News Data Generation and Augmentation

Abstract

Fake news is becoming an increasingly more significant problem in today's society, especially on social media. The fact-checking field in Data Science is becoming more and more popular as people want to solve this. However, for low-resource languages, there is not much to do without training data. In this thesis, we suggest a way to generate multilingual data from a knowledge base to prevent the problem of low resources. We will use pre-trained deep learning models, like BERT to measure the quality of the generated data. Lastly, we will discuss if the data generation improved the models and if it is a feasible strategy to generate more data.