NSLM

Code & data accompanying the paper ["Unveiling Implicit Deceptive Patterns in Multi-modal Fake News via Neuro-Symbolic Reasoning"].

Prerequisites

• python > 3.7
• CUDA > 11
• Prepare requirements: pip3 install -r requirements.txt.
• Set environment variable $PJ_HOME: export PJ_HOME=/YOUR_PATH/NSLM/.

Quick Start

• Go to folder src/.
• Run train_*.sh.

Data

• You can refer to the papers released the Weibo and Fakeddit dataset for offical data.
• Here we provide Pre-processed text data of Weibo in the data/weibo folder.

• Notes:
  • You can find the output data in the out folder specified in the config file.
  • Since the probability of the three deceptive patterns appearing at the same time in actual situations is very small, in the experiment we set the y corresponding to the situation where all three deceptive modes exist to 2 (that is, a category that does not exist in the dataset), so the decoder finally becomes a three-category prediction

Acknowledgment

Our implementation is mainly based on follows. Thanks for their authors. https://github.com/jiangjiechen/LOREN