DANES

Deep Neural Network Ensemble Architecture for Social and Textual Context-aware Fake News Detection

Article:

Ciprian-Octavian Truică, Elena-Simona Apostol, Panagiotis Karras. DANES: Deep Neural Network Ensemble Architecture for Social and Textual Context-aware Fake News Detection. Knowledge-Based Systems, 294:1-13(111715), ISSN 0950-7051, June 2024. DOI: 10.1016/j.knosys.2024.111715

Packages

Python >= 3.9

• SciPy
• Pandas
• numpy
• SciKit-Learn
• matplotlib
• tensorflow
• stop_words
• nltk
• SpaCy

Utilization

To process the text and create both word embeddings and social context embeddings use

python create_embeddings.py FILE_NAME

The FILE_NAME is a csv file with the followind columns ['id', 'content', 'label', 'num_reactions', 'num_comments', 'num_shares', 'num_likes', 'num_loves', 'num_wows', 'num_hahas', 'num_sads', 'num_angrys']. The output of this script is

• corpus.mat - the tokenize corpus
• network.mat - the social context embeddings
• w2v_cbow.mat - the Word2Vec CBWO embeddings
• w2v_sg.mat - the Word2Vec Skip-Gram embeddings
• ft_cbow.mat - the FastText CBOW embeddings
• ft_sg.mat - the FastText Skip-Gram embeddings
• glove.mat - the GloVe embeddings
• mittens.mat - the Mittens embeddings

To train the [Bi]GRU DANES vesion use

python danes_gru.py

To train the [Bi]LSTM DANES vesion use

python danes_lstm.py