Relation Extraction

Question 1

What is relation extraction?

Answer 1

Identifying relationships amongst entities in a piece of text

Question 2

what are the three main relationship types?

Answer 2

affiliations, geospatial, part of

Question 3

How can we approach relation extraction using a pattern based approach?

Answer 3

regular expressions

Question 4

how can we modify the expression /* has a hub at */ to include entity constraints?

Answer 4

/[organisation] has a hub at [location]/

Question 5

How can we reduce false negatives when using regex for relation extraction?

Answer 5

relax the pattern or extend the set of patterns using bootstrapping

Question 6

What is bootstrapping?

Answer 6

We start with a set of seed patterns and seed tuples, e.g. /* has a hub at */ and (ryanair, charleron).

We search for the tuple terms in the seed tuple

Use search results to extract new patterns

Search for additional tuples using the new patterns

Question 7

How can we perform relation extraction using a statistical/ml approach

Answer 7

• requires labelled data

a) detect the relationship
b) classify the relationship

Question 8

How do we detect the relationship using a statistical/ml approach?

Answer 8

do two entities participate in a relationship- binary classification

Question 9

How do we classify the relationship using a statistical/ml approach?

Answer 9

Multiclass classification

Question 10

How can we combine detecting the relationship and classifying the relationship into one ml task?

Answer 10

multiclass classification, where one of the labels is no relationship

Question 11

what types of features can we use in our ml approach?

Answer 11

• Named entity features: entity type, head word
• Context: words between, words before the first entity, words after the second entity (within a fixed window)
• Syntactic structure: dependency paths

Question 12

What is a dependency path?

Answer 12

Follow dependencies to create a path

e.g. easyjet at liverpool