Complexity reduction in GraphSAGE with importance based selection in homophilic arrangements

Graph Neural Networks have challenges in processing of large datasets. Although several models have attempted to reduce complexity of processing large graphs, the problem is still not fully resolved. This work attempts to provide a homophily based approach to improvised on the earlier approaches. The important-features based homophlic approach is a pre-processing technique that reduces the data size which has been observed to reduce the complexity significantly. Although the technique is only implemented in the GraphSAGE model but as it is a pre-processing technique, it can be applied to other Convolutional Graph Neural Network models also thereby improving their performance.