GAT2VEC

Representation Learning for Attributed Graphs is a framework for learning a represenation using multiple sources of information.

DIRECTORIES:

1. src :- has the source code for GAT2VEC, and evaluation
2. data :- it contains the input networks in respective directories, along with labels for classification task the files adjedges.txt, labels.txt, and docs.txt are the orignal files of the datasets. we preprocess and generate the files for GAT2VEC processing and to bring the uniformity in the for vertex id's We uniformly start vertex id's from 1.
3. embeddings: the embeddings learned are stored in this directory

DATA FORMAT GAT2VEC reads network in adjacency list. It needs two types of files:

1. _graph.adjlist : This adjacency list represents the structural graph (directed or undirected).
2. _na.adjlist: This adjacency list is an undirected bipartite graph. The structural vertices are numbered from to 1 to num. of structural nodes, and the attribute vertices are numbered after structural vertices. This bipartite graph doesn't contain labels as attributes.

The file <network_name>_label_10_na.adjlist is a bipartite graph in which labels of 10% of nodes are incorporated as attributes.

USAGE: To learn a representation without using label information.

1. #python __main__.py --data M10

To learn a representation using labels.

2. #python __main__.py --data M10 --label True

To learn a representation only using bi-partite graph

3. #python __main__.py --data M10 --algo g2vbip

Paper

Please cite our paper if you find the code useful for your research.

@inproceedings{gat2vec,
  booktitle="Journal of Computing",
	author = {Sheikh, Nasrullah and Kefato, Zekarias T. and Montresor, Alberto},
	title = "gat2vec: Representation Learning for Attributed Graphs",
	month = May,
	year = 2018,
	type = {JOURNAL},
	pdfurl = {https://link.springer.com/content/pdf/10.1007/s00607-018-0622-9.pdf},	
	publisher = {Springer},
}

PS: The pre-processing code for generating structural and bipartite graph will be uploaded soon My python code has influence of Java :)