+ Can be done in a way resembling the neural entity linking models: computing similarity between the encoding of the question text and the encoding for each possible relation.

Relation detection and linking

- Most of the algorithms in this scenario have a final stage which takes the top j
entities and the top k relations returned by the entity and relation inference steps,
searches the knowledge base for triples containing those entities and relations, and
then ranks those triples.
- Thus, results are heuristic.

Ranking of answers

+ An RDF triple is a 3-tuple, a predicate with two arguments, expressing
some simple relation or proposition.
+ Can be used to answer text questions like “When was Ada Lovelace
born?” or “Who was born in 1815?”.




Saint Olaf College

The system builds a semantic representation of the query and then use it to query databases of facts

Knowledge-based Question Answering (QA)

Knowledge-Based QA from RDF triple stores (Graph-based QA)

- A semantic parser is used to map the question to a structured program to produce an answer. These logical forms can take the form of some version of predicate calculus, a query language like SQL or SPARQL.
-  Semantic parsing algorithms can be supervised fully with questions paired with a hand-built logical form or can be weakly supervised by questions paired with an answer.

Learn Before

Related

Learn After