Code for CS4248 Group Project (AY22/23 Sem 2) [code]
By Dane Marc Yap Bagaoisan, Hu Yue, Joshua Emmanuel Teo Rui Zhong, Ni Shenghan, Sajal Vaishnav, Wang Huan
Recent advancements in generative artificial intelligence (AI) capabilities have made content generation easier than ever, exacerbating the spread of misinformation in media. As a possible solution to this issue, our research aims to develop a machine learning classifier that could help by automatically identifying news articles as reliable, satirical, hoax, or propaganda. We theorised that these categories appealed to the reader's emotions differently and thus experimented with different ways of preprocessing text, creating embeddings and enhancing those embeddings with sentiment features. We compared a number of machine learning models and found that a combination of TextCNN and LSTM models performed the best.
nlp
|-embedding/
|-models/
|-preprocessing/
embedding/contains code we used to produce the document embeddingsmodel/contains code we used to establish baselines, train, run inference, and test our modelspreprocessing/contains the python notebooks we used to create 13 preprocessing strategies
- Download the folders from this drive link
- These folders contain the various embeddings and pre-processed datasets we produced to train our models
- Each individual python file or notebook inside the repo is runnable on it's own, given the file paths are setup correctly
-
Compare to The Knowledge: Graph Neural Fake News Detection with External Knowledge https://aclanthology.org/2021.acl-long.62/ github:https://github.com/BUPT-GAMMA/CompareNet_FakeNewsDetection
-
Truth of Varying Shades: Analyzing Language in Fake News and Political Fact-Checking https://aclanthology.org/D17-1317.pdf
-
Do Sentence Interactions Matter ? Leveraging Sentence Level Representations for Fake News Classification https://arxiv.org/pdf/1910.12203.pdf github:https://github.com/MysteryVaibhav/fake_news_semantics