Co-Regularized Deep Multi-Network Embedding

This is a reference implementation for DMNE. The DMNE algorithm learns node representations on multi-network data. Please refer to the following paper for details.

Reference:

Co-Regularized Deep Multi-Network Embedding<br> Jingchao Ni, Shiyu Chang, Xiao Liu, Wei Cheng, Haifeng Chen, Dongkuan Xu and Xiang Zhang<br> Proceedings of the International Conference on World Wide Web (WWW), 2018.

For any questions about the code, please contact Jingchao Ni (jingchaoni@psu.edu).

Input

The format of the input data is the edge list of each network.

Domain network

node_id node_id edge_weight

For undirected networks, the same edge will be written in two directions, e.g., 1, 2, 1.00 and 2, 1, 1.00.

Cross-network relationship

node_id_in_domain_1 node_id_in_domain_2 relationship_weight

Label in each domain (for evaluation)

node_id label

Output

For each network, there is an output file in emb/. If a network has n nodes, there are n+1 lines in its output file. The first line contains the number of nodes and the dimensionality of the embeddings.

num_of_nodes dim_of_embedding

The next n lines contain node embeddings.

node_id dim_1 dim_2 ... dim_d

where dim_1, ..., dim_d are the d-dimensional embedding of a node.

Running

• Install libsvm in libsvm/.
• Run rundemo.m to see the demo program on 6ng dataset.

Citing

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

@inproceedings{ni2018co,
  title={Co-Regularized Deep Multi-Network Embedding},
  author={Ni, Jingchao and Chang, Shiyu and Liu, Xiao and Cheng, Wei and Chen, Haifeng and Xu, Dongkuan and Zhang, Xiang},
  booktitle={Proceedings of the International Conference on World Wide Web (WWW)},
  pages={469--478},
  year={2018},
  organization={International World Wide Web Conferences Steering Committee}
}