Fake-News-Detection

Domain-Specific Fake News Detection

Installation Instructions

1. Install latest version of CUDA if running on GPU (RECOMMENDED): https://docs.nvidia.com/cuda/cuda-installation-guide-microsoft-windows/index.html

2. Install latest version of Python: https://www.python.org/downloads/

3. Install latest version of Python pip
   curl https://bootst/rap.pypa.io/get-pip.py -o get-pip.py
python get-pip.py

4. Install jupyter notebook
   pip install jupyter

5. Clone repo locally
   git clone https://github.com/howardt12345/Fake-News-Detection.git

Installing Torch with Cuda

1. Navigate to https://pytorch.org/get-started/locally/

2. Choose the OS, package, language(Python) and CUDA Version Appropriate to your machine

3. Copy the command and run in the chosen platform terminal

4. To check CUDA version run nvcc --version

5. To check if CUDA is installed and working open Python interpreter, import torch and run torch.cuda.is_available()

To Run Data Pre-Processing

1. Navigate to Fake-News-Detection/preprocessing/<insert category/dataset _ Preprocessing>.ipynb

2. Run all cells

3. View output in the notebook and feather file in the /data/ folder in base directory

To Run Classification

1. Navigate to base directory

2. Run Python /classification/bert.py

3. View output in the /classification/logs/ directory

To Run Detection

This will train and test models

1. Navigate to Fake-News-Detection/detection/run_detection.ipynb
2. Change parameters according to anticipated results
3. Run run_detection.ipynb and wait for completion
4. View log in category folder to see results, if I run politics models, I navigate to Fake-News-Detection/results/politics and find the log file training_log_DATE_TIME.log

To analyze results
Note: there must be a valid log file to read from

1. Navigate to Fake-News-Detection/detection/result_analysis.ipynb
2. Run the notebook and view cell outputs

Proposed Architecture

The fake news detection will have the following stages:

1. Categorization stage: Categorize the input and pass the input to the model trained for the specific domain
   • https://towardsdatascience.com/topics-per-class-using-bertopic-252314f2640
2. Detection stage: With models trained to detect fake news within a specific domain, detect whether the input is fake news or not

Potential Datasets

• Politics: (both)
  • LIAR Dataset: Contains statements made by political figures, labeled by their truthfulness. (https://huggingface.co/datasets/liar)
  • ISOT Dataset (multipurpose dataset) (https://onlineacademiccommunity.uvic.ca/isot/2022/11/27/fake-news-detection-datasets/)
  • Fake News Dataset (https://www.kaggle.com/competitions/fake-news/data)
  • PolitiFact Dataset: Includes fact-checks of political statements. (https://www.kaggle.com/datasets/rmisra/politifact-fact-check-dataset) (https://github.com/KaiDMML/FakeNewsNet)
• Healthcare and Medicine: (one is enough - choose the larger)
  • HealthMisinfo Dataset: Focuses on COVID-19 misinformation in healthcare. (https://trec-health-misinfo.github.io/)
  • FakeHealth Dataset: Contains fake health-related articles. (https://github.com/EnyanDai/FakeHealth)
• Science and Technology: (one is enough - choose the larger)
  • BuzzFeedNews Dataset: Includes news articles from BuzzFeed News. (https://www.kaggle.com/code/sohamohajeri/buzzfeed-news-analysis-and-classification)
  • FakeTech: Contains technology-related fake news.
• Finance and Stock Markets: (one is enough - choose the larger)
  • Stock Market Fake News Dataset: Focuses on news related to stock markets.
  • Financial News Dataset: Contains financial news articles.
• Entertainment and Celebrity News: (one is enough - choose the larger)
  • CelebA Dataset: Contains celebrity images (for image-based fake news). (https://mmlab.ie.cuhk.edu.hk/projects/CelebA.html)
  • Entertainment News Dataset: Focuses on entertainment news.
• Sports: (one is enough - choose the larger)
  • SportFake Dataset: Contains sports-related fake news articles.
  • Sports News Dataset: Contains sports news articles.
• Environment and Climate Change: (one is enough - choose the larger)
  • Climate Change Misinformation Dataset: Focuses on climate change misinformation. (https://huggingface.co/datasets/climate_fever/viewer/default/test?row=0)
  • Environmental News Dataset: Contains news related to the environment.
• Education:
  • Education Misinformation Dataset: Focuses on educational misinformation.
• Crime and Security: (FA-KES is important)
  • Crime News Dataset: Contains news articles related to crime.
  • FA-KES (https://zenodo.org/record/2607278)
• Travel and Tourism:
  • Travel Misinformation Dataset: Focuses on travel-related misinformation.
• Religion and Spirituality:
  • Religion Misinformation Dataset: Contains religious misinformation. (https://data.mendeley.com/datasets/5ykks3psks/5)
• Transportation:
  • Transportation Misinformation Dataset: Focuses on transportation-related misinformation.
• Social Media and Online Discussions:
  • PHEME (https://figshare.com/articles/dataset/PHEME_rumour_scheme_dataset_journalism_use_case/2068650)
  • GossipCop (https://github.com/KaiDMML/FakeNewsNet)
  • Fake News Detection (https://www.kaggle.com/jruvika/fake-news-detection/version/1)