HPHG

The code and data for CIKM '19 paper "Hyper-Path-Based Representation Learning for Hyper-Networks".

Readers are welcomed to star/fork this repository to reproduce the experiments and train your own model. If you find this code useful, please kindly cite our paper:

@inproceedings{huang2019hyper,
  title={Hyper-Path-Based Representation Learning for Hyper-Networks},
  author={Huang, Jie and Liu, Xin and Song, Yangqiu},
  booktitle={CIKM},
  year={2019}
}

Examples

Train embeddings (HPHG)

python src/main.py --input graph/GPS/train.edgelist --output emb/GPS/HPHG.emb --save-model model/GPS/model.h5 --alpha 100 --iter 5

python src/main.py --input graph/Drugs/train.edgelist --output emb/Drugs/HPHG.emb --save-model model/Drugs/model.h5 --alpha 100 --iter 1

Train embeddings (HPSG)

python src/main.py --input graph/GPS/train.edgelist --output emb/GPS/HPSG.emb --alpha 100 --iter 15 --method hpsg

python src/main.py --input graph/Drugs/train.edgelist --output emb/Drugs/HPSG.emb --alpha 100 --iter 5 --method hpsg

Link prediction (HPHG)

python src/main.py --input graph/GPS/train.edgelist --load-model model/GPS/model.h5

Link prediction (HPSG)

python src/link_prediction.py --input emb/GPS/HPSG.emb

Options

You can check out the other options available to use with HPHG or HPSG using:

python src/main.py --help

Input

The supported input format is an edgelist:

1            # 1 for heterogeneous and 0 for homogeneous
146 70 5     # number of nodes of each node type (heterogeneous)
0 93 203 220 # edge_name node_id node_id node_id
1 17 153 216
...

Output

The output file has n+1 lines for a graph with n nodes. The first line has the following format:

num_of_nodes dim_of_representation

The next n lines are as follows: node_name dim1 dim2 ... dimd

where dim1, ... , dimd is the d-dimensional representation learned by HPHG or HPSG.

Misc

• Please store the edgelist file in graph/<dataset>/* and store the embedding file in emb/<dataset>/*.
• This implementation is only applied to 3-uniform heterogeneous hyper-networks, if you have any question or need a compatible version, you are welcome to open an issue or send me an email.