FakeNewsMitigation

A modified Neural Diffusion Model based on user profile embedding included users profile basic information and profile tweets text embedding(by using the SBERT attention mechanism) with the goal of cascade prediction and fake news mitigation in social networks.

Project Structure

This project includes user embedding and model parts. In the user embedding directory, “cascade_embedding.py” includes methods for creating users embeddings with input datasets and save user embeddings via pickle file. The “cluster_text.py” file includes methods for clustering users vectors and saves clusters information via pickle file. Finally, “vis.py” includes a method for visualizing users clusters. Due to the large volume of the source datasets, we don’t push these files in this project. Datasets are accessible via Lab Servers. For output example, there are “clusters_vectors.p” and “low_dim” files in this part of the project.

In Model Directory, the training phase is done via the “train.py” file, and the Datasets address should be set via the “DataLoader_Feat.py” file. Evaluating model is via “evaluate_new.py” file and for some thesis diagrams, there is “rus_script.py” file which is used as the script for making different diagrams.

For reviewing more source codes NearalDiffuseModel is accessible via Lab Servers in the Rafiee directory; also NDM project source code is available via GitHub.