OntologyDriven Automatic Entity Disambiguation in Unstructured Text

1
Ontology-Driven Automatic Entity Disambiguation
inUnstructured Text

• Joseph Hassell, Boanerges Aleman-Meza I. Budak
  Arpinar
• Presented by
• Abhelaksh Thakur
  

2
What is the paper about?

• Problem Entity Disambiguation
• How to exploit background information for entity
  disambiguation?
• Aim
• To create a method to disambiguate entities
  within unstructured text by using clues in the
  text and exploiting metadata from an ontology.
• Also to provide an implementation of the method
  using a very large, real world ontology.
• Dataset used
• Approach
• Algorithm
• Evaluation of the algorithm (Precision Recall)
• Related Work
• Future work

3
TERMS USED

• Entity disambiguation
• Ontology
• Semantic web
• RDF
• DBLP
• DBWorld

4
DATASET

• DBLP Digital Bibliography Library Project
• Contains bibliographic information for computer
  science researchers, journals and proceedings.
• it indexes more than 725,000 articles and
  contains a few thousand links to home pages of
  computer scientists.
• It provides two XML files
• First contains objects such as authors,
  proceedings and journals.
• Second contains lists of papers.
• The first XML file was converted into an RDF
  containing 3,079,414 entities.

5
DATASET(cont.)

• DBWorld
• DBWorld is a mailing list of information for
  upcoming conferences related to the databases
  field.
• An HTML scraper was created that visits the
  DBWorld site and downloads only the posts that
  contain Call for Papers, Call for
  Participation or CFP in the subject
• The system disambiguates the people listed in
  these postings and provides a URI to the
  corresponding entity in the ontology.

6
APPROACH

• Aim To find the correct entity out of the
  various possible matches.

7
APPROACH (cont.)

• Scenario Disambiguating researchers by their
  names appearing in the DBWorld postings.
• Spotting entity names.
• Text proximity relationships.
• Text Co-occurrence Relationships.
• Popular Entities.
• Semantic Relationships.

8
(No Transcript)
9
ENTITY DISAMBIGUATION ALGORITHM- Terms Used

• es confidence score
• cf initial confidence score
• acf initial abbreviated confidence score
• pr proximity score
• co text occurrence score
• sr semantic relationship score
• pe popular entity score

10
ENTITY DISAMBIGUATION ALGORITHM

• Step 1 Spot Entity Names
• es cf / no. of entities with
  
• same label
• es acf / no. of related entities in
• the ontology
• acf feature is used for abbreviated names and
  can be turned on or off.

11

• Step 2 Spot literal value of Text-proximity
  relationships
• es espr if Text-proximity relationship found
  near candidate entity

12

• Step 3 Spot literal value of Text-co occurrence
  relationships
• es esco if text co occurrence relationship
  found.
• Step 4 Using popular entities
• es espe for those entities that have many
  relationships that were predefined as popular.

13

• Step 5 Using semantic relationships
• esessr
• a) Choose candidate entity with highest es and
  increase es of entities that have semantic
  relationships (eg. co-authorship) in ontology
  with it.
• b) Ignore candidate entity with low es.

14

• Step 5 (contd.)
• Threshold es es(t)
• Iterate till atleast one candidate entity gets
  its es increased over es(t).

15
ENTITY DISAMBIGUATION ALGORITHM

• Algorithm Disambiguation( )
• for (each entity in ontology)
• if (entity found in document)
• create candidate entity
• es for candidate entity lt- cf / (entities in
  ontology)
• for (each candidate entity)
• search for candidate entitys text proximity
  relationship
• if (text proximity relationship found near
  candidate entity)
• es for candidate entity lt- es for candidate
  entity pr
• search for candidate entitys text
  co-occurrence relationship
• if (text co-occurrence relationship found)
• es for candidate entity lt- es for candidate
  entity co
• if (ten or more popular entity relationships
  exist)
• es for candidate entity lt- es for candidate
  entity pe

16
SAMPLE OUTPUT OF THE ALGORITHM
17
EVALUATION OF THE ALGORITHM

• Disambiguated dataset (A)
• Set of entities found by our algo (B)
• Precision size of (A n B)/size of (B)
• Recall size of ( A n B)/size of (A)

18
INPUT SETTINGS

• Description Variable
  Value
• charOffset
  50
• Text proximity relationships pr
  50
• Text co-occurrence relationships co
  10
• Popular entity score pe
  10
• Semantic relationship sr
  20
• Initial confidence score cf
  90
• Initial abbreviated confidence
• score acf
  70
• Threshold threshold
  90

19
RESULTS

• Correct Disambiguation 602(A n B)
• Found Entities 620 (B)
• Total Entities 758 (A)
• Precision 97.1
• Recall 79.4

20
MEASURE OF PRECISION RECALL ON PER-DOCUMENT
BASIS
21
CHARACTERISTICS OF THE ALGORITHM

• Performs well on unstructured text
• Reduces need for string similarity computations.
• Does not require training data

22
RELATED WORK

• KIM
• Automatic ontology population system.
• Natural language processor/ indexing.
• SCORE
• Semantic metadata management system
• Uses associations from knowledgebase.
• It is a commercial system.
• ESpotter
• Uses a lexicon and/or patterns to recognize named
  entities.

23
FUTURE WORK

• Integration of results of entity disambiguation
  into a more robust platform like UIMA
  (Unstructured Information Management Architecture)

24
THANK YOU..!!