I want to learn the relationship between 2 nodes, say a User U and a location L. I have a small graph network representing the different interactions between U and L. For example, U is connected to property P, which is in turn connected to L.

A graph G possesses the following properties:

  1. ~10-15 nodes
  2. Heterogenous nodes of 3-5 categories with varying trust associated (needs to be learnt)
  3. Heterogeneous edges with different confidences (needs to be learnt)
  4. Upto 10 possible paths between U and L, paths can vary in length from 1-6.

Given a binary labelled set of such graphs representing whether U is connected L , I need to determine confidence of Connection(U, L). This would involve learning the confidence associated with each node and edge and then, the confidence of connection (U, L).

What is a good way to model this data? One way I can think of is to extract signals from all nodes/edges and train a regression model. Most graph techniques work well for large graphs with mostly homogenous nodes. Would there be any techniques I can employ here?

  • $\begingroup$ Do you have many small graphs or just one? $\endgroup$ – roman Jul 2 '20 at 21:34

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.