Fake News Detection

Fake News Detection - A Comparative Study of Deep Learning Computational Methods

Advanced Machine Learning Project - A.Y. 2023/24

Team:

• Mattia Piazzalunga - 851931
• Nicolò Urbani - 856213

Benchmark Datasets:

• WELFake dataset - Verma et al.
• ISOT dataset - Ahmed et al.

Abstract

”Fake news” are a central issue in today’s society, a phenomenon intensifid by the advent of social media which has facilitated their dissemination. Automatically detecting them, therefore, becomes a fundamental challenge involving research. In this paper, the state-of- the-art has been achieved on two benchmark datasets for ”fake news detection”: ISOT and WELFake, achieving 100% accuracy in classi- fying news from the former and 97% from the latter. Additionally, a custom neural network was built, which, with a 99.96% parameter reduction, achieved accuracy close to the most performant model stud- ied in this analysis, based on BERT. Back Translation and Dataset Combination proved unhelpful techniques in attempting to improve model generalization for this type of task.

Requirements

Before executing the Python code, make sure you have the following packages installed:

pip install contractions
pip install spacy
pip install tensorflow
pip install keras
pip install inflect
pip install -q -U tensorflow-text
pip install -q tf-models-official

Setting Directory Base Path

In the code section 'Main global variables & functions' set the base path directory in where you have stored the project files:

python path_to_drive_folder = "/content/drive/MyDrive/Advanced_Machine_Learning"

Change this path according to your directory structure.

Necessary Folders

Ensure the following folders exist within your project directory:

datasets
preprocessed_dataset
models
txt

Running the Code

After setting up the directory structure and installing the required packages, you can run the Python code provided.