Update

An early version of this code has a bug that makes it incorrectly processes undirected graphs. This bug is rectified in the current version.

Layout

1. data
2. algos
3. eval
4. embds

Input

See an example in data/wiki/

1. attrs.txt
2. edgelist.txt
3. labels.txt
4. full.mat

Download edgelists & labels from here Download mat files from here

Output

In embds/wiki/

Preprocessing

1. Split training, testing and negative edge sets for link prediction

$ cd eval/
$ python splitTrainTest.py --action split --data wiki --ratio=0.3
$ python splitTrainTest.py --action select --data wiki --ratio=0.3

Then convert edgelist.train.txt to a train.mat, which contains four variables: 'A', 'P', 'Dout', 'Din' and format '-v7.3'. 'A' is adjacency matrix, 'P' is transition matrix, 'Dout' is out-degree array, 'Din' is in-degree array. 'A' and 'P' are sparse matrices.

2. Generate node pairs for graph reconstruction

$ cd eval/
$ python gen_nodepairs.py --data wiki --ratio=0.01

Algorithm

See readme.md in algos/nrp/

Evaluation

$ cd eval/
$ python eval_linkpred.py --algo nrp --data wiki --d 128
$ python graphreconstruct_util.py --algo nrp --data wiki --d 128

Citation

@article{yang2020homogeneous,
  title={Homogeneous network embedding for massive graphs via reweighted personalized PageRank},
  author={Yang, Renchi and Shi, Jieming and Xiao, Xiaokui and Yang, Yin and Bhowmick, Sourav S},
  journal={Proceedings of the VLDB Endowment},
  volume={13},
  number={5},
  pages={670--683},
  year={2020},
  publisher={VLDB Endowment}
}