search

neo4j / graph-data-science

Source code for the Neo4j Graph Data Science library of graph algorithms.
https://neo4j.com/docs/graph-data-science/current/
Other
621 stars 160 forks source link

Directed link prediction #163

Open devineyfajr opened 2 years ago

devineyfajr commented 2 years ago

Is your feature request related to a problem? Please describe.

Addresses a limitation of the current method.

Describe the solution you would like

Possibly, a non-commutative feature combiner, say concatenation of the source node and target node feature vectors.

Describe alternatives you have considered

Additional context

AliciaFrame commented 2 years ago

@devineyfajr thanks for adding this issue.

I think this feature request is really for directed link prediction - having our ML pipelines be aware of the direction of relationships in the training data, and making directed predictions in the output (eg. A -[:CAUSES]-> B but not B-[:CAUSES]->A).

This is a super relevant ask, but requires a fair amount of work to implement. We've put this in our backlog for a future release, and we'll update you as soon as we pick this up!

If you can share more about your use case or what you're looking for in particular, that will be helpful in future scoping.

devineyfajr commented 2 years ago

Yes, directed link prediction Without getting too specific wrt use cases, think Identity graphs, where the two identities involved in a relationship have different roles in the relationship. Say A IS_MOTHER_OF_SON B. currently, we predict the link, then apply rules to determine which node is the mother and which the son. Or, A IS_EMPLOYEE_OF B. Actually, we use a generic type, IS_RELATED_TO, with edge properties to describe the roles, but the problem is the same.