README.rst

DeepWalk

DeepWalk uses short random walks to learn representations for vertices in graphs.

Usage

Example Usage: $deepwalk --input example_graphs/karate.adjlist --output karate.embeddings

--input: input_filename

--format adjlist for an adjacency list, e.g:
1 2 3 4 5 6 7 8 9 11 12 13 14 18 20 22 32
2 1 3 4 8 14 18 20 22 31
3 1 2 4 8 9 10 14 28 29 33
...
--format edgelist for an edge list, e.g:
1 2
1 3
1 4
...
--format mat for a Matlab .mat file containing an adjacency matrix

(note, you must also specify the variable name of the adjacency matrix --matfile-variable-name)

--output: output_filename

The output representations in skipgram format - first line is header, all other lines are node-id and d dimensional representation:
34 64
1 0.016579 -0.033659 0.342167 -0.046998 ...
2 -0.007003 0.265891 -0.351422 0.043923 ...
...

Full Command List: The full list of command line options is available with $deepwalk --help

Evaluation

Here, we will show how to evaluate DeepWalk on the BlogCatalog dataset used in the DeepWalk paper. First, we run the following command to produce its DeepWalk embeddings:

deepwalk --format mat --input example_graphs/blogcatalog.mat
--max-memory-data-size 0 --number-walks 80 --representation-size 128 --walk-length 40 --window-size 10
--workers 1 --output example_graphs/blogcatalog.embeddings

The parameters specified here are the same as in the paper. If you are using a multi-core machine, try to set --workers to a larger number for faster training. On a single machine with 24 Xeon E5-2620 @ 2.00GHz CPUs, this command takes about 20 minutes to finish (--workers is set to 20). Then, we evaluate the learned embeddings on a multi-label node classification task with example_graphs/scoring.py:

python example_graphs/scoring.py --emb example_graphs/blogcatalog.embeddings
--network example_graphs/blogcatalog.mat
--num-shuffle 10 --all

This command finishes in 8 minutes on the same machine. For faster evaluation, you can set --num-shuffle to a smaller number, but expect more fluctuation in performance. The micro F1 and macro F1 scores we get with different ratio of labeled nodes are as follows:

% Labeled Nodes	10%	20%	30%	40%	50%	60%	70%	80%	90%
Micro-F1 (%)	35.86	38.51	39.96	40.76	41.51	41.85	42.27	42.35	42.40
Macro-F1 (%)	21.08	23.98	25.71	26.73	27.68	28.28	28.88	28.70	28.21

Note that the current version of DeepWalk is based on a newer version of gensim, which may have a different implementation of the word2vec model. To completely reproduce the results in our paper, you will probably have to install an older version of gensim(version 0.10.2).

Requirements

numpy
scipy

(may have to be independently installed) or pip install -r requirements.txt to install all dependencies

Installation

cd deepwalk
pip install -r requirements.txt
python setup.py install

Citing

If you find DeepWalk useful in your research, we ask that you cite the following paper:

@inproceedings{Perozzi:2014:DOL:2623330.2623732,
 author = {Perozzi, Bryan and Al-Rfou, Rami and Skiena, Steven},
 title = {DeepWalk: Online Learning of Social Representations},
 booktitle = {Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining},
 series = {KDD '14},
 year = {2014},
 isbn = {978-1-4503-2956-9},
 location = {New York, New York, USA},
 pages = {701--710},
 numpages = {10},
 url = {http://doi.acm.org/10.1145/2623330.2623732},
 doi = {10.1145/2623330.2623732},
 acmid = {2623732},
 publisher = {ACM},
 address = {New York, NY, USA},
 keywords = {deep learning, latent representations, learning with partial labels, network classification, online learning, social networks},
}