Named Entity Recognition

Question 1

what is a named entity

Answer 1

a real world object that can be named, e.g. person, location, time, money, organisation

Question 2

How long is an entity mention

Answer 2

can be either a single token or a span of text

Question 3

What are the possible approaches to named entity recognition

Answer 3

• dictionary lookup
• rule based
• machine learning

Question 4

How can we treat named entity recognition as a ml problem

Answer 4

treat it as a tagging problem

use BIO for entity mention and then find the category

Question 5

How many classes do we consider with n entity types

Answer 5

2n + 1

Question 6

What is the local approach to NER

Answer 6

tags are independent of each other

Question 7

What methods are local approaches to NER

Answer 7

rnn, lstm, bilstm

Question 8

What does a global approach to NER mean

Answer 8

tags are dependent on each other

Question 9

What methods are global approaches to NER

Answer 9

hmm (hidden markov model)

crf (conditional random fields)

Question 10

How does a HMM perform NER

Answer 10

establish sequence by arranging output variables in a chain

sequence input x, sequence of states y

yt depends only on yt-1
xt depends only on yt

Question 11

Give the equation for sequences P(y,x) using a HMM

Answer 11

p(y,x) = multiply for each t p(yt-1|yt)p(xt|yt)

Question 12

What is a conditional random field crf?

Answer 12

a discriminative model for sequence labelling

finds the most probable sequence y’ given observation sequence x

Question 13

What is the equation for y* using a crf model

Answer 13

y* = argmax p(y|x)

Question 14

what is p(y|x) given a linear chain crf

Answer 14

1/normalisation factor *

exp( sum for t sum for feature f weight * feature function(yt, yt-1, xt)

Question 15

What is a feature function?

Answer 15

characterises the input (based on certain features)

Question 16

what feature types are there (7)

Answer 16

• contextual
• POS tag
• trigger words
• length in tokens
• orthographic (captials, punct, single char)
• suffixes
• gazetteers features (features from a list)

Question 17

Give an example feature function

f(yt, yt-1, xt) = …

Answer 17

1 if first letter of xt is uppercase

0 otherwise

Question 18

The more features…

Answer 18

the more powerful the learner

Question 19

What are the benefits of crf (3)

Answer 19

• features are intuitive
• easy to interpret and debug
• high performance

Question 20

what is the disadvantage of crf

Answer 20

feature engineering requires domain knowledge

Question 21

what is the solution to the feature engineering requirements of crf

Answer 21

neural networks

Question 22

why are neural networks the solution to the feature engineering requirements of crf

Answer 22

we can represent word meanings in a high dimension space that shows the features of the word

Question 23

How can we use a rnn for NER

Answer 23

use softmax on an rnn to predict the tag given a token. use a training set of tagged sentences

Question 24

What is the equation we minimise for rnn for NER

Answer 24

negative log likelihood:

sum each training sentence: sum each tag: - log p(tag|token)

Question 25

In the global approach, for a given sequence we predict…

Answer 25

all tags, a whole token sequence

Question 26

how can we implement a global approach for NER using crf

Answer 26

use a linear chain crf but replace the feature function with whi + b from a bilstm

Question 27

why is a global approach better than local for sequence labelling (2)

Answer 27

we can encode rules like in BIO, I never comes after O

In a local approach, more Os is bad but it doesn’t affect the global approach

Question 28

Benefits of of CRF for NER (4)

Answer 28

feature engineering
no pretrained vectors
interpretable
performs well with many NE categories

Question 29

Key features of NN (4)

Answer 29

dont need features
need pretrained vectors from large models
not easy to interpret
performs less well when lots of NE categories