Implementation of our KDD21 paper Causal Understanding of Fake News Dissemination on Social Media [1]
- Run the script user_attribute.py to preprocess data.
- Run the script create_bipartite.py to creat the user-news bipartite graph.
- To get the News- and User-News-based propensity score estimation, run the script pscore.py and pscore_ut.py, respectively.
- 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.
- Note that the main programs (BPRMF.py or NCF.py) mostly are adapted from code for paper Neural Graph Collaborative Filtering.
- python == 3.7
- tensorflow == 1.14.0
[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.