The implementation of gScarf proposed in IJCAI'19. If you use our software, please cite the following paper.

Hiroaki Shiokawa, Toshiyuki Amagasa, Hiroyuki Kitagawa, 
"Scaling Fine-grained Modularity Clustering for Massive Graphs,"
In Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI2019), pp.4597-4604, Macao, China, August 2019. 

@inproceedings{ShiokawaTK2019,
  title     = {{Scaling Fine-grained Modularity Clustering for Massive Graphs}},
  author    = {Shiokawa, Hiroaki and Amagasa, Toshiyuki and Kitagawa, Hiroyuki},
  booktitle = {Proceedings of the 28th International Joint Conference on
               Artificial Intelligence, {IJCAI-19}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},             
  pages     = {4597--4604},
  year      = {2019},
  month     = {7},
  doi       = {10.24963/ijcai.2019/639},
  url       = {https://doi.org/10.24963/ijcai.2019/639},
}

This software is released under the MIT license. Please read LICENSE.txt for details.

gScarf requires the following softwares.

• gcc Version 4.8.5 (or later)
• boost C++ Libraries Version 1.53.0 (or later)

We have confirmed that our software works on the following environments.

• CentOS 7.6
• MacOS 10.14.6

1. Install gcc and boost libraries on your environment.
2. Rewrite Makefile. You should add your file path of "libboost_serialization.a" in the directory of the boost libraries to Makefile like as follows:

LIB= (add your path of "libboost_serialization.a" here)

3. Build the software.

$ make clean; make

4. If you can find gscarf and convert, the build has been successfully completed.

Input file must be formatted as a list of edges included in a given graph; each line represents a pair of node IDs that are connected via an edge. The nodes in the same line must be spanned by TAB or space like as follows:

1	90
1	109
1	207
1	282
1	321
1	699
...

gscarf reads the given graph by the CRS format, and this requires a file conversion process. To covert the input file (sample_graph.txt) into the CRS format (sample_graph.bin), you should run convert like as follows:

$ ./convert -i sample_graph.txt -o sample_graph.bin

convert requires two options, -i and -o, that specify names of the input file and the CRS formatted file, respectively.

Note that convert will converts the input file as an undirected graph in this file conversion process even though the input file represents a directed graph.

Finally, we can run the clustering by gscarf like as follows:

$ ./gscarf sample_graph.bin

gscarf has the following two options.

• Display the clustering result; each line indicates a pair of node ID and corresponding cluster ID. For example, in the following example, nodes 991 and 993 are included in a cluster 21.

$ ./gscarf -v sample_graph.bin
...
991	21
992	8
993	21
994	8
995	4
996	4
997	8
998	17
999	4
1000	4

• You can display statistics of the clustering result by using option -d.

$ ./gscarf sample_graph.bin -d
Number of nodes: 1001
Number of edges: 15534
Runtime for clustering (sec.):0
Modularity Q:0.851665
Number of clusters:36
Average cluster size:27.8056
Largest cluster size:49

• Hiroaki Shiokawa, Toshiyuki Amagasa, Hiroyuki Kitagawa, "Scaling Fine-grained Modularity Clustering for Massive Graphs," In Proceedings of the 28th International Joint Conference on Artificial Intelligence (IJCAI2019), pp.4597-4604, Macao, China, August 2019.
• http://www.kde.cs.tsukuba.ac.jp/~shion/index.html.en