Automatic Keyphrase Extraction

1
Automatic Keyphrase Extraction

• Emma Nguyen
• Wing group meeting
• December 15th 2006

2
Outline

• Related work
• Method
• Evaluation
• Future work

3
Related work

• Approaches
• Keyphrase assignment
• Predefined list of potential keyphrases -gt
  document
• Disadvantage human expertise
• Keyphrase extraction
• Document -gt list of candidate keyphrases -gt
  output keyphrases (based on set of features)

4
Keyphrase extraction

• Frank et al 1999 KEA
• TFxIDF
• Position of phrases first occurrence
• Turney 2003
• Enhanced KEA
• Pointwise Mutual Information
• Disadvantage time and resource consuming

5
Method

• Based on KEA
• Only noun phrases (POS tagger)
• Features
• TFxIDF
• Number of words in phrase
• Acronym
• Section related
• Number of sections appear
• Position in section
• Frequency in section

6
Section Finder

• Abstract
• Introduction
• Related work
• Method
• Evaluation
• Reference

7
Features

• Number of sections
• A Number of sections phrase P appears
• B Number of total sections in document
• Position in section
• C The first sentence in section S that contains
  phrase P
• D Number of total sentences in section S
• Frequency in section

8
Evaluation

• n number of matches between the output phrases
  and author-assigned phrases
• N number of total extracted keyphrases

Training document 50
Testing document 53
Precision 16.2
9
Evaluation (Cont)
Author-assigned keyphrases Extracted keyphrases
Dispersers Ramsey graphs Independent sources Extractors disperser sources independent sources extractor Result
Implicit feedback Personalized search User model Interactive retrieval Search engine Information retrieval User model Web search Implicit user
10
Future work

• Revise the added features
• Study the collocation of words