Submitted by **J00Nnn** t3_zlmvcp
in **MachineLearning**

I am solving a graph-level problem. I want to fit graph embeddings to a learn-to-rank NN to rank the graphs by their "quality". The "quality" of the graphs is determined by whether they have certain shape or structure, say they have self-loops and no loose end, has many split nodes and merging nodes etc. The node and edge features are not in consideration.

To my understanding, graph embeddings are best suited for graph similarity comparison, are there any techniques that can fit my use case?

Weary-Marionberry-15t1_j06hbfh wroteIs your “quality” known for each graph? If so, I recon you could do this by simply building a convolutional graph neural network and phrase this as a regression problem (assuming quality is a float).