Causal-Understanding-of-Fake-News-Dissemination

Implementation of our KDD21 paper Causal Understanding of Fake News Dissemination on Social Media [1]

Code usage

1. Run the script user_attribute.py to preprocess data.
2. Run the script create_bipartite.py to creat the user-news bipartite graph.
3. To get the News- and User-News-based propensity score estimation, run the script pscore.py and pscore_ut.py, respectively.
4. For the method BPRMF, simply run BPRMF.py, BPRMF_t.py, BPRMF_ut.py, and BPRMF_neural.py. They correspond to the biased model and unbiased models using news-, user-news-, and neural-network-based propensity score estimations. This also applies to the method NCF.
5. Note that the main programs (BPRMF.py or NCF.py) mostly are adapted from code for paper Neural Graph Collaborative Filtering.

Python packages version

• python == 3.7
• tensorflow == 1.14.0

Reference

[1] Lu Cheng, Ruocheng Guo, Kai Shu and Huan Liu. Causal Understanding of Fake News Dissemination on Social Media. ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), 2021.