Relation Extraction
- Tags
- text-processing
Identify the relationships between pairs of (pre-identified) entities. Eg. "Both these organisations are in the same town", "this organ supports this body part", "FOO is an employee in ORG".
We [see page 7, define] R as the set of possible (assumed binary) relations between entities (which're a subset of those identified in the entity extraction stage).
This task is decomposed into:
| Task | Meaning |
|---|---|
| Relation detection | Find pairs of entities between which a relation holds. |
| Relation classification | Determine the type of a previously detected relation. |
[see page 11, Challenges]
Expressive Form - There're many different ways to express the same relation, this includes:
- canonical. states the relationship exactly "foo works-at bar"
- synonyms. canonical but using alternative words with same meaning "foo employed-at bar"
- syntactic variations (structure variations). "bar employs a lot of people including foo"
Coreferencing - Texts often contain several co-referenced terms and you need to unpack them before you can extract relations.
Information can be spread over [see page 11, multiple] sentences and may be implied rather than explicitly asserted.