Improve country linking

[willu47]

A naive implementation of linking outputs to countries has been implemented based on the CONTAINS operand.

MATCH (c:Country)
CALL {
  WITH c
  MATCH (o:Output)
  WHERE o.abstract CONTAINS c.name
  AND NOT exists((o:Output)-[:REFERS_TO]->(c:Country))
  CREATE (o)-[r:REFERS_TO]->(c)
  RETURN r
}
RETURN r

This results in country names (such as Niger) being linked when the abstract contains a country name (e.g. Nigeria CONTAINS Niger; South Sudan CONTAINS Sudan etc.).

The linking should be improved so that full country names are linked.

• One approach would be to use entity extraction using a Hugginface transformer-based LLM.
• Another would be to force matching of whole words, rather than substrings, so Niger != Nigeria