NER + Dependency Parsing

I am trying to figure out how I can utilize the spaCy dependency tree + NER and/or pattern matching to associate characteristics for entities I am identifying in my text. For example, "I really like red motorcycles with black saddle bags".

I want to be able to identify "red motorcycles" as an entity (I can do this part) but also associate the black saddle bags. Is there a way with the dependency tree (other) to do that? I can leverage the matchers / NER models to get the entities but the associated characteristics is unclear. Any help would be greatly appreciated.

It would be great if I could somehow create the association

  • Entity: red motorcycle
  • Characteristics: black saddle bags

A subset of my matcher rule to get the entities:

matcher.add("VEHICLE", None,
            [colors,vehicles],
            [vehicles],
            [colors,{"LOWER":"car"}],
            [{"LOWER":"car"}]
           )

What I am ultimately looking for is to see if I can generate or pull out the below.

VEHICLE - red motorcycles ATTRIBUTES - black saddle bags

Is using NER + Dependency Parsing the way to pursue this output?

Check out this related question + answer: Annotation scheme / workflow for entity relations

1 Like