Application and analysis of embedding methods for node classification in homophily-rich networks
Tan, Yee Ying
Date of Issue2019
School of Physical and Mathematical Sciences
Using the structural link information in homophily-rich network graphs can potentially improve node classification accuracies. This research will test out the feasibility of using embedding methods to embed the structural link information in the homophily-rich network graphs for the purpose of facilitating node classification tasks. In particular, embedding similar nodes closer together and dissimilar nodes further apart. This should help increase class separability and classification accuracy in homophily-rich network graphs. The three embedding methods used in this research are Multidimensional Scaling (MDS), Laplacian Eigenmaps (LE) and GraRep. Results showed high classification accuracies when the structural link information was incorporated. This shows the importance of structural link information for node classification in homophily-rich network graphs and the success of the embedding methods.
Final Year Project (FYP)