Semi-automatic Entity-Relationship Modelling Through Natural Language Processing Nazlia Omar Supervisors: Dr. Paul Hanna, Prof. Paul Mc Kevitt Faculty of Engineering University of Ulster, Jordanstown

1
Semi-automatic Entity-Relationship Modelling
Through Natural Language ProcessingNazlia
OmarSupervisorsDr. Paul Hanna, Prof. Paul Mc
Kevitt Faculty of EngineeringUniversity of
Ulster, Jordanstown 
2
Objectives of research

• Design and implement ER-Converter to transform
  natural language specifications of database
  problems into Entity-Relationship (ER) models
• Develop new heuristics to assist transformation
• Evaluate approach against human performance and
  compare to other work in area

3
Previous work

• E-R Generator (Gomez et al., 1999)
• Dialogue Tool (RADD) (Buccholz et al., 1995)
• DMG (Tjoa and Berger, 1993)
• ANNAPURA (Eick and Lockemann, 1985)
• CM-Builder (Harmain and Gaizauskas, 2003)

4
Heuristics in Database Design

• Heuristics are simple procedures, often guided
  by common sense, that are used to provide, easily
  and quickly, good but not necessarily optimal
  solutions to difficult problems.
• (Zanakis and Evans, 1981, p. 84)
• To determine entities, attributes, relationships
  and cardinalities
• Gathered from past work and newly formed

5
Example Heuristics

• HE1- A common noun may indicate an entity type.
• HE2- A proper noun may indicate an entity.
• HE3- A gerund (noun converted from a verb or
  known as verbal noun) may indicate an entity
  type which is converted from a relationship type.
• HE4- If two consecutive nouns are present, check
  the second noun. If it is not one of these words
  (number, no,id, address and name), most likely it
  is an entity. Else it may indicate an attribute.
  

6
Heuristics Weights

• Each heuristic assigned weight depending on level
  of confidence
• Example
• HE6gtweight gt "0.60",elementgt "Entity",status
  gt "New",,
• HA2gtweight gt "-0.50",elementgt
  "Attribute",status gt "New",,

7
Architecture of ER-Converter
 
Natural Language Requirements Specification
8
ER-Converter

• Step 1 Read natural language input text into
  Memory-based Shallow Parser
• 2 Part-of-speech tagging
• 3 Remove plurals
• 4 Apply heuristics
• 5 Assign weights
• 6 Human intervention
• 7 Produce final ER model

9
Evaluation Measures

N
Recall

• Other measures overgenerated, undergenerated,
  ask user, unattached and wrongly attached

correct
N
N

correct
missing
N

correct
Precision

N
N
incorrect
correct
10
Experimental Results

• Based on 30 natural language specifications
• Evaluation results

11
Frequency Distribution
12
Frequency Distribution
13
Frequency Distribution
14
Comparison with Related Work
15
Conclusion and Future Work

• Formation of NEW heuristics show contribution as
  supported by evaluation results
• Integration of WordNet
• Semantic analysis