This repository provides an implementation of the method proposed in "FraudNE: a Joint Embedding Approach for Fraud Detection", Mengyu Zheng, Chuan Zhou, Jia Wu, Shirui Pan, Jinqiao Shi and Li Guo, IJCNN 2018

• input/ contains an example graphs 'zomato.edgelist.400';
• output/ is the directory to store the learned node embeddings;
• src/ contains the implementation of the proposed FraudNE method.

The implementation is tested under Pyrhon 2.7, with the folowing packages installed:

• networkx==1.11
• numpy==1.11.2
• tensorflow==1.5

The code takes a bipartite input graph composed users and items. Every row indicates an edge between two nodes, such like:

          user_node1_id_int item_node2_id_int weight_int

The file does not contain a header. Nodes can be indexed starting with any non-negative number.

The graph is assumed to be directed and weighted by default.

If you find FraudNE useful for your research, please consider citing the following paper:

                @inproceedings{zheng2018fraudne,
                  title={Fraudne: a joint embedding approach for fraud detection}, 
                  author={Zheng, Mengyu and Zhou, Chuan and Wu, Jia and Pan, Shirui and Shi, Jinqiao and Guo, Li},  
                  booktitle={2018 International Joint Conference on Neural Networks (IJCNN)}, 
                  pages={1--8}, 
                  year={2018},
                  organization={IEEE}}