ASiNE: Adversarial Signed Network Embedding

This repository provides a reference implementation of ASiNE as described in the following paper:

ASiNE: Adversarial Signed Network Embedding
Yeon-Chang Lee, Nayoun Seo, Kyungsik Han and Sang-Wook Kim
43rd Int'l ACM Conf. on Research and Development in Information Retrieval (ACM SIGIR 2020)

Authors

• Yeon-Chang Lee (lyc0324@hanyang.ac.kr)
• Nayun Seo (nayounseo@hanyang.ac.kr)
• Kyungsik Han (kyungsikhan@ajou.ac.kr)
• Sang-Wook Kim (wook@hanyang.ac.kr)

Input

The input files should be saved in data/ folder. The structure of the input file is the following:

node_id1 node_id2 sign

Node ids start from 0 to N-1 (N is the number of nodes in the graph).

Output

The output files are saved in results/ folder. The first line has the following format:

num_of_nodes dim_of_embeddings

The next N lines are as follows:

node_id dim1 dim2 ... dimk

where dim1, ... , dimk is the k-dimensional embedding vectors learned by ASiNE.

Arguments

--dataset                 Dataset name (default: "wikirfa")
--n_emb                   Dimensionality of embedding (default: 128)
--lr                      Learning rate (default: 0.01)
--window_size             Size of context window (default: 2)
--learn_fake_pos          Whether performing the additional learning of the fake positive edges generated from negative generator (default: False)
--n_epochs                Number of epochs to train (default: 20)
--n_epochs_gen            Number of iterations for the generator per epoch (default: 10)
--n_epochs_dis            Number of iterations for the discriminator per epoch (default: 10)
--n_sample_gen            Number of generating edges (default: 20)
--batch_size_gen          Batch size for the generator (default: 64)
--batch_size_dis          Batch size for the discriminator (default: 64)
--n_node_subsets          Number of subsets to divide nodes existing in the positive or negative graph for large datasets (default: 1)
--lambda_gen              L2 loss regularization weight for the generator (default: 0.00001)
--lambda_dis              L2 loss regularization weight for the discriminator (default: 0.00001)    

Basic Usage

python ./src/main.py --dataset wikirfa --n_emb 128 --lr 0.01 --window_size 2 --learn_fake_pos True --n_epochs 20 --n_node_subsets 2  

Requirements

The code has been tested running under Python 3.6. The required packages are as follows:

• tensorflow == 1.8.0
• numpy == 1.14.5
• pandas == 0.24.2
• scikit-learn == 0.19.1
• tqdm == 4.23.4

Cite

We encourage you to cite our paper if you have used the code in your work. You can use the following BibTex citation:

@inproceedings{lee20sigir,
  author   = {Yeon{-}Chang Lee and Nayoun Seo and Kyungsik Han and Sang{-}Wook Kim},
  title     = {ASiNE: Adversarial Signed Network Embedding},
  booktitle = {International ACM SIGIR Conference on Research and Development in Information Retrieval (ACM SIGIR 2020)},      
  year      = {2020}
}