IRJET- Automatic Database Schema Generator

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 11 | Nov 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 3598
Automatic Database Schema Generator
Sayali Sant1, Amruthkala Bhat2, Neha Tiwari3, Purva Raut4
1,2,3Student, Department of Information Technology, Dwarkadas J. Sanghvi College of Engineering, Mumbai, India
4Assistant Professor, Department of Information Technology, Dwarkadas J. Sanghvi College of Engineering,
Mumbai, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Automatic Database Schema Generator is a tool
that facilitates schema designing from Natural language
based textual requirements as an input from the user, thus,
automating the process of extracting probable entities and
their attributes, identifying primary and foreign keys, etc. and
thereby, eliminating time consuming requirement analysis
phase in the project developmentlifecycle. Thispaperproposes
how Natural language based textual requirements can be
analysed to extract the entities from plain text and produce a
database schema. Natural language text can tend to be
ambiguous due to varied contextualmeaningsassociatedwith
the words. In order to reduce this ambiguity, thesystemmakes
use of Domain Ontology to identify associated terms in the
given context, thus increasing theefficiencyandrichnessof the
database hence created.
Key Words: Domain ontology, Natural Language
Processing, Schema, Entities, Attributes.
1. INTRODUCTION
A majority of projects in the industry make use of user data
or other related information that needs to be stored in the
application systems for further use. A large amount of such
information might be later used by the application system
itself for further processes or by the service provider. All
these linked data are saved in application databasesforlater
use. Therefore, designing a database that includes majority
of useful information and that does not save redundant or
unimportant data is crucial. Currently this process iscarried
out manually by repeated analysis of client requirements
and multiple iterations ofvalidationfromthe clients.Amajor
chunk of Software development time and efforts are to be
invested in the initial phases of the cycle for the projecttobe
a success. It is important that the fundamental idea of the
project cycle be clear and strong enough. Any loophole in
this phase will lead to a cascading effect in further stages.
Generating the Database Schema for the project from user
specifications involves multiple iterations of requirement
analysis and rigorous client communications for validation.
Projects of all scales make use of this approach irrespective
of the type of solution or repetition of business uses. Rapid
Application Development processes involve short spanned
cycles, where schema generation from analysis result is
important, failure of which mayaffectthetimelydeployment
of the product.
The system aims to create an automatedenvironmentfor
analysing textual user requirements and formulating a
relevant and client-domain specific database schema by
extracting appropriate entities, associated attributes from
the analysed text using Domain Ontology. The entities are
then mapped according to their relations and key attribute
values are identified to compose a full-fledged Relational
Schema for the end user.
2. RELATED WORK
2.1. Automatic generation of schema from nested
key- value data
This system automaticallytransformsself-describing, nested
key-value data formats such as JSON, commonly found in
NoSQL systems into traditional relational data that can be
stored in standard RDBMS. [3] This process includes a
schema generation algorithm that discovers relationships
across attributes of deformalized datasets to organize them
into relational tables. The next process includes a matching
algorithm to identify attributes with overlapping entities to
merge them together under one entity to reduce data
repetition. This system is most useful in cases where
databases need to be propagated from a NoSQL system to a
relational system thus helping users gain semantic
understanding of complex, nested datasets.
Figure -1: JSON conversion to RDBMS format
Performance:
For a large Twitter dataset, the three consecutive phases of
the process required 31 hours, 3.6 hours, 31 minutes
respectively. In the second phase, it was observed that some

attribute matches were semantically related; however, a
large number of matches were not semantically related and
should not have been matched.
2.2. Automatic Relational Schema Extraction from
Natural Language RequirementsSpecification Text
[1] This methodology deals with automatic construction of
the relational database schema by identifying the key
attributes from SRS using“rule-basedapproach”.Thesystem
architecture and working of each module in the system is as
below:
Input: The system takes natural language textual
requirements as input.
Domain Knowledge Elicitor: [1] This module splits the SRS
into sentences and tags each word from all the sentences as
nouns, verbs, adjectives, etc. using POS (Parts of Speech)
tagging. Tagging of words is necessary to chunk the words
that form noun phrases or verb phrases. The phrases are
classified using simple phrasal grammar.
Schema Generator: This module identifies entities,
attributes, methods and relationships based on simple rule-
based approach from S-V-O pattern: Translating Nouns to
Entities, Translating Noun-Noun to entity property
according to the position, Translating the lexical verb of a
non-personal noun toa methodofthisnoun,TranslatingS-V-
O structure to a class diagram with Subject and Object as
Entity and verb as relation.
Identification of Primary Key (PK) and Foreign Key (FK):
[1]A rule based approach is used to identify the primary key
attribute from the attributes of all the tables.
3. PROPOSED APPROACH TO BUILD
Now we discuss the detailed approach of our proposed
system to extract entities and their respective attributes
from natural language requirements. It includes mainly
following modules. [2] They are accepting the text input and
tokenizing it followed by their POS tagging, parsing of
ontology represented in OWL, identification of entities and
attributes and finally, extraction of key attributes [1].
First module outputs the tagged text using a “Parts of
Speech” tagger (POS) from which nouns, noun phrases, and
verbs can be identified. Next module is OWL parser which
parses the ontology represented in OWL and thus the
classes, ObjectProperties and DataProperties are extracted.
Now the nouns and noun phrases become the candidate for
entities and attributes identification. In the last two steps
domain ontology is used to explore the important concepts
of the domain. Identificationofentitiesandtheirattributesis
followed by extraction of key attributesfortheentities.Thus,
at the end we get the desired relational database schema.
Figure -2: Architecture of the proposed system
Following section describes each module in detail:
POS tagger: POS tagger i.e. Part Of Speech tagger will parse
the entire text document of user requirements and will tag
each word according to its part of speech (e.g. Noun, verb,
adjective, adverb, etc.) and extract all the verbs and nouns
from it. Now nouns serve as a source of attribute and class
identification.
OWL Parser: It parses the extracted nouns and verbs using
OWL parser. In OWL i.e. Ontology Web Language, domain
specific conceptual terms are termed as entities,
relationships are termed as ObjectProperty and attributes
are represented as DataProperty. So, with the help of OWL
Parser these are extracted and stored.
Entities and Attributes extractor: Itextractsthecomponents
like entities and attributes from the nouns and noun-
phrases. Domain ontology is used here to extract contextual
entities and attributes. Attributes are specified as
DataProperty in OWL. If domain ontology does not contain
any information about the nouns and noun phrases under
consideration, then the tagged nouns and noun phrases are
again processed and semantic similarity is taken into
consideration in such cases.
Key attribute Extractor: A rule-based approach is used to
identify the primary key attribute from the attributes of all

the tables. Following this, anotherrulebasedapproachhelps
in retrieving foreign keys and their associated entities.
4. IMPLEMENTATION
The system is implemented using Python Language.
The input for the system is Problem statement and Problem
domain. The problem statement is in the form of natural text
which is then Tokenized, Tagged and Lemmatized using
spaCy[8]. spaCy is an open-source software library for
advanced Natural Language Processing, written in the
programming languages Python and Cython. Based on the
input problem domain a suitable Domain Ontology is
webscraped and parsed using libraries such as
BeautifulSoup and OntoSpy. Nouns from thetaggedproblem
statement are classified as Entities and attributes based on
lexical and semantic similarity with those extracted after
parsing the OWL file that was webscraped based on suitable
input Domain ontology. To tackle the ambiguity caused due
to use of Natural Language in problem statement and to
overcome lexical dissimilarity of common contextual words,
the system makes use of Google Dictionary to locate
probable synonyms in the problem statement. The schema
obtained is then marked with PrimarykeysandForeignkeys
according to the relationship between the entities withrule-
based algorithms that is explained in detail in the following
paragraph. The system aims to provide user flexibility and
thus recommends them few extra appropriate and probable
entities and attributes that are fit for the problem at hand
and can be included in the Schema on need basis. An extra
functionality provided by our tool is retrieval of relational
database schema if the user wishes to only submit the
domain name or theme of the project and not the entire
customized problem statement. In this scenario, we simply
parse the OWL file of the respective Domain Ontology and
provide with the relevant classes, attributes, key attributes
and recommended schema related elements.
Rule-based Algorithm for determining Primary keys: -
FOR EACH attribute IN attribute list of an entity
1. Find attributes with substring
“_no/_number/_ID” and String_1=Split and
extract the word lying before
“_no/_number/_ID”
2. IF String_1 matches with Entity name, THEN it
qualifies to be the primary key.
3. ELSE IF Any of the Meanings/Synonyms (that
are retrieved through Google dictionary) of
String_1 matches with the entity name, THEN it
qualifies to be the primary key.
4. ELSE no primary key exists and the entity
qualifies to be a weak entity.
Rule-based Algorithm for determining Foreign keys: -
FOR EACH attribute IN attribute list of an entity
1. Find attributes with substring “_no/_number/_ID”
and String_1=Split and extract thewordlyingbefore
“_no/_number/_ID”
2. IF String_1 does not match with the Entity name(E)
and matches with one of the Entity name in the
Entity list(L) THEN it qualifies to be theforeign key
of E with the parent Entity being the matchedEntity
name from L.
3. ELSE IF Any of the Meanings/Synonyms(that are
retrieved through Google dictionary) of String_1
does not match with the Entity name(E) and
matches with one of the Entity name in the Entity
list(L) THEN it qualifies to be the foreign key of E
with the parent Entity being the matched Entity
name from L
4. ELSE IF String_1 does not match with the Entity
name(E) and matches with the Synonyms of one of
the Entity names in the Entity list(L) THEN it
qualifies to be the foreign key of E with the parent
Entity being the matched Entity name from L
5. ELSE no foreign key exists.
Let us consider an example of an “Hospital Management
System”. The problem domain entered by the user is
“Hospital” and the problem statement entered by the user
which is in the natural language based textual format is as
follows: -
“Patients request forappointmentforanydoctorbyspecifying
the doctor_name. and doctor_id. The details of the existing
patients like Name, Address, age and gender are retrieved by
the system. New patients update their details in the system
using a unique patient_id allotted to them before they request
for appointment with the help of assistant_name where each
assistant can be distinguished based on their distinct
Assistant_id. The assistant confirms the appointmentbasedon
the availability of free slots for the respective Medical
practitioners and the patient isinformed. Assistantmaycancel
the appointment at any time.”
Firstly, the input text is tokenized, tagged and lemmatized.
Secondly, “hospital.owl” i.e. an Ontology based on user
entered problem domain is retrieved using web scraping.
Thereby, the nouns such as Patients, Doctor, Assistant, slots,
etc. are classified as entities and Name, Address, age and
gender, doctor_name, etc. areclassifiedasattributeswith the
help of the ontology. Ambiguous words like “Doctors” and
“Medical Practitioners” are handled further to avoid
redundant entities and attributes. The primary key and
foreign key of each entity are identified using appropriate
rule-based approach. Thus, a database schema is generated.

The expected output is as follows: -
Figure -3: Automatic Database Schema Generator can
generate schema from customised problem statement
provided by user or from the domain type of user
requirement. (For e.g.: User may give a problem statement
for Library system as input or only specify ‘Library’ as
domain)
Figure -4: The textual input and domain of the problem
statement is provided as input.
Figure -5: The system generates a schema for the given
problem statement and identifies Primary keys and
Foreign keys.
Figure -6: The system also provides some probable
schema elements that can be included along with the ones
specified by the user to enhance the database.
5. CONCLUSION
The Automatic Database Schema Generator tool has been
successfully implemented for extracting database schema
from natural language textual input problem statement.The
tool can analyze words with similar context and integrates
the results accordingly. The keys have been identified to
denote referential integrity and uniquely identify the
entities. An entire repository of commonly used Ontologies
proved to be favorable for mapping various attributes to
their respective entities. Thus, a well enriched database
schema is retrieved with the help of this tool.
REFERENCES
[1] Automatic Relational Schema Extraction
from Natural LanguageRequirementsSpecificationText -
S. Geetha and G.S. Anandha Mala, JNTU Hyderabad
Andhra Pradesh, India, Head / CSE, St. Joseph’sCollegeof
Engineering, India, IDOSI Publications, 2014
DOI: 10.5829/idosi.mejsr.2014.21.03.21475
Patients Doctor Assistant
Name Doctor_name Assistant_name
Gender Specialization Degree
Age Degree Address
Address Address Age
Age
Patient_id
(Primary
key)
Doctor_id
(Primary key)
Assistant_id
(Primary key)

[2] Domain Ontology Based Class Diagram Generation From
Functional Requirements Jyothilakshmi M S and Philip
Samuel Information Technology, School of Engineering,
Cochin University of Science and Technology, Kochi-22
Kerala India International Journal of Computer
Information Systems and Industrial Management
Applications. ISSN 2150-7988 Volume 6 (2014) pp. 227
[3] Automatic Generation of Normalized Relational Schemas
from Nested Key-Value Data Michael DiScala, Yale
University Daniel J. Abadi, Yale University
[4] Class diagram extraction from textual requirements
using Natural language processing (NLP) techniques
Mohd Ibrahim Al-Qaoud, Rodina Ahmad Department of
Software Engineering, Faculty of Computer Science and
Information technology, University of Malaya, Malaysia
[5] Auto-generation of Class Diagram from Free-text
Functional Specifications and Domain Ontology
Xiaohua Zhou, Nan Zhou
[6] NLP based Object Oriented Analysis and Design from
Requirement Specification Subhash K.Shinde LT College
of Engg. Navi Mumbai Varunakshi Bhojane PIIT New
Panvel, Navi Mumbai International Journal of Computer
Applications (0975 – 8887) Volume 47– No.21, June
2012.
[7] Natural Language ProcessingElizabethD.LiddySyracuse
University, liddy@syr.edu [8] Fensel D. (2001)
Ontologies. In: Ontologies. Springer, Berlin, Heidelberg,
Print ISBN 978-3-662-04398-1
[8] spaCy:https://spacy.io/Websites
https://protege.stanford.edu/publications/ontology_dev
elop ment/ontology101-noy-mcguinness.html

IRJET- Automatic Database Schema Generator

More Related Content

What's hot (17)

Similar to IRJET- Automatic Database Schema Generator (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET- Automatic Database Schema Generator