You are on page 1of 1

Community detection using Graph Convolutional Neural Network

OBJECTIVE
Communities represent similar interests, similar opinions, similar functions. Complex
networks are gained much attention with the advancement of the internet in the owing to their
prominence and popularity in recent years. Community detection is a significant and
tremendously beneficial implementation in both technical and data study. However, relevance
of classic community identification approaches such as a spectral community and statistical
inference, are decreasing as deep learning algorithms reveal a growing ability with excellent
performance to handle high-dimensional data.
DESCRIPTION
For community detection I had used the Cora dataset which is used extensively in the research
area. From the previous research paper they had used normal adjacency matrix as a input to the
Graph Convolutional layer and they got only 0.53 accuracy. So I had used to reconstruct the
adjacency matrix to group the neighbor nodes as a sub matrix. By reconstructing the adjacency
matrix there is a certain order in the adjacency matrix. Later the adjacency matrix is a high
dimensional data. So I had encoded or graph embedding to convert the high dimensional data
to low dimensional data for smoothening my resources. The input layer contains nodes which
are equal to the number of nodes in Graph. The output layer is same input layer. In the hidden
layers we had used ReLu activation function. By doing reconstructed adjacency matrix instead
of normal adjacency matrix the accuracy increases from 0.53 to 0.92 which is turned to best
algorithm in the GCN research area.
TEAM
STUDENTS
Ankem Gopi Chand, Integrated Btech and Mtech, CSE
FACULTY
Dhanpratap ingh, Assistant Professor, LPU

You might also like