This is a python implementation of the paper "GraRep: Learning Graph Representations with Global Structural Information". This project improves the implementation of the paper, applies the idea on a new datasets(Facebook Social Circles), implements a K-means algorithm, does clustering on the datasets and visulises the result using TSNE tool.
-
The original implementation needs a preprocessing of filering the isolated nodes. However, the isolated nodes can be necessary in some application. So, This implementation takes isolated nodes into consideration. With changes as following: remove the isolated nodes in preprocessing step
-
The original paper factorises the matrix using SVD to prove that the learned ditributted learning is improved by the global structural information. This project generates a distributted representation using Autoencoder instead of Matrix Factorization.
-
This project applies the improved implementation to the datasets (Facebook Social Circles). The learned distribution is used to clustering. The applied Clustring algorithm is K-means, in which the cosine distance function is used. This project also visulise the culstering result using TSNE tool.