Named Entity Recognition

1
Named Entity Recognition

• Mitch Marcus
• Slides adapted from
• Lance Ramshaw, BBN Technologies
• Salim Roukos, IBM Research (not distributed)
• Elizabeth Boschee, Michael Levit, Marjorie
  Freedman, BBN Technologies (not distributed)

2
Review Named Entities

• The who, where, when how much in a sentence
• The task identify atomic elements of information
  in text
• person names
• company/organization names
• locations
• datestimes
• percentages
• monetary amounts

3
Extraction Example

• George Garrick, 40 years old, president of the
  London-based European Information Services Inc.,
  was appointed chief executive officer ofNielsen
  Marketing Research, USA.

George Garrick, 40 years old,
Nielsen Marketing Research, USA.
4
Review Applications

• Information Extraction
• Summary generation
• Machine Translation
• Document organization/classification
• Automatic indexing of books
• Improve Internet search results(location
  Clinton/South Carolina vs. PresidentClinton)

5
Machine Learning Approaches

• ML approaches frequently break down the NE task
  into two parts
• Find entity boundaries
• Classify entities into NE categories
• Or Reduce NE boundary detection and
  classification to IOB tagging
• O outside, B-XXX first word in NE, I-XXX
  all other words in NE
• Argentina B-LOCplayed Owith ODel B-PERBos
  que I-PER

6
IdentiFinder (Nymble) Bikel et al 99

• Based on Hidden Markov Models
• Features
• Capitalisation
• Numeric symbols
• Punctuation marks
• Position in the sentence
• 14 features in total, combining above info, e.g.,
  containsDigitAndDash (e.g. 09-96),
  containsDigitAndComma (e.g. 23,000.00)

7
Nymbles structure (simplified)
START_OF_SENTENCE
END_OF_SENTENCE
PERSON
ORGANIZATION
5 other name classes
NOT_A_NAME
8
IdentiFinder (2)

• MUC-6 (English) and MET-1(Spanish) corpora used
  for evaluation
• Mixed case English
• IdentiFinder - 94.9 f-measure
• Best rule-based 96.4
• Spanish mixed case
• IdentiFinder 90
• Best rule-based - 93
• Lower case names, noisy training data, less
  training data
• Training data 650,000 words, but similar
  performance with half of the data. Less than
  100,000 words reduce the performance to below 90
  on English

9
(No Transcript)
10
(No Transcript)
11
(No Transcript)
12
one out of date but simple method
13
NER as a benchmark task

• CONLL (Conference on Natural Language Learning)
  shared task 2002, 2003
• Two languages, English German
• Following Information from
• Introduction to the CoNLL-2003 Shared Task
  Language-Independent Named Entity Recognition
• Erik F. Tjong Kim Sang and Fien De Meulder

14
Data Sets Annotation
15
Feature Representations Used
16
Methods Used

• Max Entropy Bender, ChieuNg, CurranClark
• Max Entropy others Florianal, Kleinal,
• HMMs Florianal, Kleinal, Mayfieldal,
  WhitelawPatrick
• CRFs Kleinal
• WINNOW ZhangJohnson
• Voted Perceptron Carrerasal
• RNN Hammerton
• TBL Florianal
• SVM Mayfieldal

17
Evaluation
18
Additional Data System Availability Matters