AI Based UML Diagrams Generator

K. A. D Oshada Kasun Kiringoda Arachchi

2021
AI Based UML Diagrams Generator

A dissertation submitted for the Degree of Master of

Computer Science

K. A. D Oshada Kasun Kiringoda Arachchi

University of Colombo School of Computing
2021
 DECLARATION
I hereby declare that the thesis is my original work and it has been written by me in its entirety.
I have duly acknowledged all the sources of information which have been used in the thesis.
This thesis has also not been submitted for any degree in any university previously.

Student Name: K. A. D Oshada Kasun Kiringoda Arachchi

Registration Number: 2018/MCS/047

Index Number: 18440474

_________________________________30/11/2021____

Signature of the Student & Date

This is to certify that this thesis is based on the work of Mr. /Ms. K. A. D Oshada Kasun
Kiringoda Arachchi under my supervision. The thesis has been prepared according to the format
stipulated and is of acceptable standard.

Certified by,

Supervisor Name: Prof G.K.A. Dias

__________________________

Signature of the Supervisor & Date 30/11/2021

i
I would like to dedicate this thesis To My Beloved Parents….

ii
 ACKNOWLEDGEMENTS

It gives me great desire to acknowledge all those who supported me during the course of this
project.

First of all, my heartfelt gratitude goes out to Prof. Kapila Dias, my project supervisor and my
project proposal approver for his support throughout the project.

To my loving parents and brother, for supporting and encouraging me to make this project a
success.

To all the academic and non-academic staff at UCSC for their support, for allowing to
continuing studies without any issue during the pandemic situation.

My colleagues at Yukon Software, for the corporation extended and the inspiration from the
beginning.

But last not least all my colleagues from UCSC, who have always been there for me,
encouraging me when I was down and strengthening me.

A very big and whole-hearted thank you to all of you. This would not have been possible if not
for all of you.

iii
 ABSTRACT
Unified Modeling Language (UML) diagrams are very useful to represent the business
requirement of any proposed system and help to design a system from the end user's point of
view. In the software analysis process, UML diagrams are drawn separately after gathering
requirement, and the time wastage of drawing these diagrams is high with using current drawing
tools due to the complexity of the business situation or the technical capabilities of the UML
diagrams. Automated UML Diagram generation tools necessity can be identified with amount
of time spent on requirement analysis and low quality of human analysis. The objective of this
project is to provide an approach to generating UML use case and class diagrams from the
functional requirement texts using natural language processing and machine learning.
Finding the key terms for class or use case diagram is reflected a classification task in machine
learning. The stories narrate different attributes of a diagram and the task is to identify key terms
for respective attributes. The use case diagram and class diagram have different set of attributes.
classification model can be designed and pre-process the data for extracting the key terms such
as actors, use case and classes from the requirement text. Furthermore, other text features like
position and distance of text can be integrated to improve accuracy. To Identify Use case and
class relationship sequence to sequence RNN model has introduced and it extracted key phrase
or relation phrase among text, which also outperformed the conventional technique
significantly. Improved strategies were proposed on this approach by substituting a sequence-
to-sequence RNNs with conventional techniques.
The proposed system is capable of providing solutions to generating usecase and class diagram
from the functional requirement text with using NLP and ML techniques. Diagram generating
has used Plant UML tool with its plan text language where user interaction also enabled.
Developed prototype is able to identify important actors, use cases and its relations with
relationship type for use case diagrams. And also for class diagrams important classes and it’s
relationships can identify. If this project is domain specific most of element would have been
identified. The developed final model has identified and generated the use case diagram and
class diagram up to a considerable extent according to the given requirement text which gives
a good idea about the business scenario.

Keywords: UML Diagram, Use case Diagram, Class Diagram, Natural Language Processing,
Machine Learning, Classification, RNN, sequence-to-sequence RNN, Plant UML

iv
 TABLE OF CONTENTS

DECLARATION ..................................................................................................................i

ACKNOWLEDGEMENTS .............................................................................................. iii

ABSTRACT .......................................................................................................................iv

TABLE OF CONTENTS .................................................................................................... v

LIST OF FIGURES .......................................................................................................... vii

LIST OF TABLES .......................................................................................................... viii

CHAPTER 1 INTRODUCTION ........................................................................................ 1

1.1 Motivation .................................................................................................................... 1

1.2 Statement of the problem ............................................................................................. 1
1.3 Research Aims and Objectives .................................................................................... 2
1.3.1 Aim ....................................................................................................................... 2
1.3.2 Objectives ............................................................................................................. 2
1.4 Scope ............................................................................................................................ 2
1.5 Structure of the Thesis ................................................................................................. 3

CHAPTER 2 LITERATURE REVIEW ............................................................................. 5

2.1 Literature Review ........................................................................................................ 5

2.1.1 Natural Language Processing and Machine Learning .......................................... 5
2.1.2 Derive Use case elements from requirement text ................................................. 6
2.1.3 Derive Class Diagram elements from requirement text ....................................... 8
2.1.4 Generate use case and class diagram .................................................................... 8
2.2 Presentation of Scientific Material .............................................................................. 9

CHAPTER 3 PROBLEM ANALYSIS AND METHODOLOGY .................................. 15

3.1 Problem Analysis ....................................................................................................... 15

3.2 Proposing Model/design ............................................................................................ 16
3.3 Methodology .............................................................................................................. 17
3.4 Algorithmic design .................................................................................................... 19
3.4.1 Preprocess Functional requirement texts ............................................................ 22
3.4.2 Machine Learning Techniques to Identify Diagram Elements ........................... 23
v
 3.5 Data acquiring methods ............................................................................................. 26

CHAPTER 4 IMPLEMENTATION ................................................................................. 28

4.1 Identifying Use case and Class Diagram Entities with NLP and ML techniques .. 28
4.2 Identifying Use case and Class Diagram Relationships with NLP and ML
techniques ......................................................................................................................... 32
4.3 Draw diagram with identified diagram elements ................................................... 36

CHAPTER 5 EVALUATION AND RESULTS ............................................................... 39

5.1 Collecting Dataset ...................................................................................................... 39

5.2 Dataset Preparation .................................................................................................... 41
5.3 Evaluation approach .................................................................................................. 42
5.4 Evaluation and Results............................................................................................... 44
5.4.1 Identifying use case diagram element from requirement text ............................ 44
5.4.2 Identifying class diagram element from requirement text .................................. 52
5.4.3 Generating usecase and class Diagram ............................................................... 58

CHAPTER 6 CONCLUSION AND FUTURE WORK .................................................... 65

6.1 Conclusion ................................................................................................................. 65

6.2 Future Work ............................................................................................................... 67

REFERENCES .................................................................................................................... I

Appendix A ........................................................................................................................ V

vi
 LIST OF FIGURES

Figure 2.1 ML Model prediction flow ....................................................................................... 9

Figure 3.1 Proposed Solution Design ....................................................................................... 17

Figure 3.2 Process flow with NLP And ML Model ................................................................. 19

Figure 4.1 Sample JSON output format ................................................................................... 36

Figure 5.1 Sample dataset format with functional requirement ............................................... 41

Figure 5.2 Confusion matrix for the case of binary classification ........................................... 43

Figure 5.3 Program generated language code for usecase diagram ......................................... 59

Figure 5.4 Edited language code for usecase diagram ............................................................. 59

Figure 5.5 Actual usecase diagram according to the functional text ........................................ 60

Figure 5.6 Generated usecase diagram using Plant UML tool ................................................. 61

Figure 5.7 Program generated language code for class diagram .............................................. 62

Figure 5.8 Edited language code for class diagram .................................................................. 62

Figure 5.9 Actual usecase diagram according to the functional text ........................................ 63

Figure 5.10 Generated class diagram using Plant UML tool ................................................... 63

vii
 LIST OF TABLES

Table 2.1 Research gaps on deriving use case diagram elements .............................................. 9
Table 2.2 Research gaps on deriving class diagram elements .................................................. 10
Table 2.3 Existing research evaluation ..................................................................................... 11
Table 3.1 Sample dataset……………………………………………………………………. 20

Table 5.1 Test data snippet for actor in functional text ............................................................ 45
Table 5.2 Model Predicted results snippet for identified actors...…………………………… 45

Table 5.3 Evaluation metric result for actors ………..……………………………………… 46

Table 5.4 Test data snippet for usecase in functional text …………………..……………… 46

Table 5.5 Model Predicted results snippet for identified usecases ........................................... 47
Table 5.6 Evaluation metric result for usecases ....................................................................... 47
Table 5.7 Data snippet from preprocessing stage data separation ............................................ 48
Table 5.8 Test data snippet for relationship involved entities in functional text...................... 49
Table 5.9 Model Predicted results snippet for identified usecase relationship involved entities
.................................................................................................................................................. 50
Table 5.10 Test data snippet for relationship Type in functional text ...................................... 51
Table 5.11 Model Predicted results snippet for identified usecase relationship type............... 51
Table 5.12 Test data snippet for classes in functional text…………………..……………… 52

Table 5.13 Model Predicted results snippet for identified classes ........................................... 53
Table 5.14 Evaluation metric result for classes… .................................................................. 53
Table 5.15 Data snippet from preprocessing stage class relationship data ............................ 54
Table 5.16 Test data snippet for class relationship involved entities in functional text ......... 55
Table 5.17 Model Predicted results snippet for identified class relationship involved entities ...
................................................................................................................................................ 56
Table 5.18 Test data snippet for class relationship type in functional text ............................ 57
Table 5.19 Model Predicted results snippet for identified class relationship type ................. 58

viii
 CHAPTER 1
INTRODUCTION
In the Software development industry Requirement gathering and Design phase plays an
important role in software development lifecycle (Whitney, E., CODE Magazine 2020).
Depending on the time taken to these phases whole project duration will be affected since
different parties who are playing different kind of role need to provide their feedback and also
some authorities need to be agreed whether the Design would be satisfied with the requirements.
More importantly particular client needs to be satisfied with the Design with gathered
requirements.

1.1 Motivation

Creating UML diagrams is a vitally important and time-consuming task which both
requirements and design phases in software Development. These diagrams like use-case
Diagram and class diagram will consider as a transition between the two phases. Use case
diagram consider as the one of the most used functional modeling techniques that use in the
software development process and the main core of Object-Oriented analysis and design where
other models are derived is class diagram.
From the initial survey carried out by Madanayake R. S., (2019) in his research stated that
“Class diagrams, ER Diagrams, User stories and use case models were the most popular
diagrams used as well as where the most Duplication of Work occurred, according to the data
analysis conducted by the researchers.”
Since most of stakeholders are not aware on this diagram technique and not have capability of
understanding the particular diagram with comparing to the business requirement may lead to
consume more time on creating these diagrams against business requirements. Hence, reducing
the time taken to create the use-case diagram against the business requirements is the motivation
behind this project.

1.2 Statement of the problem

When drawing these UML Diagrams Requirement engineer or Business analyst need to put
lot of work with gathering and analyzing business requirements and also good amount of time
will waste due to this time taken process. UML Diagrams will draw separately after gathering
requirements and wastage of time will increase when drawing these Diagram using current
diagram tools according to particular business requirement due to complexity of business
scenario or technical capability of UML diagram concepts. Automated UML Diagram
1
generating tools necessity can be identified with the how much time would spend on
analyzing business requirements and also the low quality of the human analysis.

1.3 Research Aims and Objectives

This project proposes an approach to facilitate use case and class diagram extraction from
textual requirements using AI techniques.

1.3.1 Aim

The very first phase of the software development life-cycle is Requirement engineering.
This phase, the requirements should be able to translated from the client language to the
Developer’s language. Usecase diagram and class diagram are two different approaches
used to describe the functional requirements. Since the requirements can be written in
natural language, Natural Language Processing Techniques (NLP) and Machine
learning (ML) techniques would be able used to extract the information. This proposed
model aiming to provide an approach to generate the UML usecase diagrams and class
diagrams from the requirements text provided, using natural language processing and
Machine learning.

1.3.2 Objectives

Following are the objectives that have achieved over the span of the research project:
• Identification and analysis most popular UML techniques and investigation of the
existing techniques, practices and tools about drawing UML use case and class
diagrams.
• Investigation of Natural language processing and Machine learning in the context of
identifying use case and class diagram elements.
• Development of an AI based prototype for generate of use case and class diagram by
analyzing the requirements.
• Identification of an appropriate evaluation mechanism to evaluate the solution.

1.4 Scope

This study will mainly focus on providing solution to generating Usecase diagram and
Class diagram against the particular business requirement. system will read and
understand the business requirement using Natural Language Processing and Machine
2
 learning and identifying entities and relationships on that for generate use case diagram.
And also, classes and relationship between classes will also be identified to generate
class diagram. Then system will show generated diagram with including user interacting
feature where user can add additional element and add changes to use case or class
diagram or edit existing element and its attribute or relations. There after user can
confirm to generate both use case and class diagram respectively. The proposed system
is limited only in reducing the time taken to draw use-case and class diagrams against
the business requirements in software developments projects.

1.5 Structure of the Thesis

Literature review
With this chapter our aim to identifies, evaluates and synthesizes the relevant
literature within a UML modeling, Natural Language Processing and Machine
Learning field of research. It will illuminate how UML modelling and Artificial
Intelligent techniques knowledge has evolved within the field as highlighting
what has already been done and what is emerging, what is generally accepted
and also what is the current state of thinking on the topic of UML modelling. In
addition to these literature review chapter will identify a research gap such as
under-researched or unexplored areas and articulates how this kind of research
project would addresses this gap.

Methodology
In this chapter need to answer these questions as how did I do this research and
why did I do it that way. This will cover not only the methods used to collect
and analyze relevant business requirement data, but also the theoretical approach
and relevant techniques, framework used in the term of Natural Language
Processing and Machine learning field. This would inform both the choice of
methods and the approach to interpreting those data towards generating UML
use case diagram and class diagram.

Evaluation and results

In this chapter the results of the research project are presented and discussed
evaluation with reference to the aim of the study, which was to generating UML
diagrams against Business Requirements Using Natural language processing

3
and Machine learning. The two sub-aims - the first to find the most appropriate
NLP and ML techniques towards identifying use case and class diagrams
elements from business requirement, and the second to generate diagram using
the diagram generating tool form the main comparisons in the evaluation.
Evaluation will be done using a case study and solution given in a text book with
the project output and also evaluation can be done with the help of domain
experts in order to evaluate the efficacy of the provided solution.

Conclusion and future works

This chapter would present the conclusions of the research described in this
project thesis. The aim and objectives of this research, will be outlined in the
Introduction chapter, and those are reviewed and their achievement addressed.
It summarizes the focus of the Natural language processing and Machine
learning approach on generating UML diagrams against Business
Requirements, and draws conclusions from the discussions and results in the
previous chapters. Proposals for future work indicated by the research are
suggested.

4
 CHAPTER 2
LITERATURE REVIEW

In this chapter our aim is to identifies, evaluates and synthesizes the relevant literature within a
UML modeling, Natural Language Processing and Machine Learning field of research. It will
illuminate how UML modelling and Artificial Intelligent techniques knowledge has evolved
within the field as highlighting what has already been done and what is emerging, what is
generally accepted and also what is the current state of thinking on the topic of UML modelling.
In addition to these literature review chapter will identify a research gap such as under-
researched or unexplored areas and articulates how this kind of research project would
addresses this gap.

2.1 Literature Review

In this section will explore how UML modeling evolved with Natural Language Processing
and Machine Learning field of research with identifying existing research that has been done
so far.

2.1.1 Natural Language Processing and Machine Learning

In recent past years, different approaches have studied for the natural language requirements
transformation into UML diagrams, from that few researchers have specifically focused on both
usecase and class diagram using the Natural language processing techniques along with
machine learning techniques. User requirement analysis can be an information extraction
application of Natural Language Processing. It will be identified specific semantic elements in
the business requirements entered in textual form as mentioned in Bhagat, S., Kapadni, P.,
Kapadnis, N., Patil, D. and Baheti, M. (2012). Therefore, this proposed system will come under
the information extraction over the Information Retrieval and also in Questioning and
Answering Tasks in NLP Moldovan, Dan & Surdeanu, Mihai. (2002).
In natural language processing several approaches were used to extract the information from
given text in Barba, P., Lexalytics, (2020).
• Sentence splitting
With this approach all text will split into sentences.
• Lexical Analysis
This would get the split sentences and tokenize the sentences into words.
• Syntax analysis
5
 This approach will receive the tokens generated by lexical analysis and applies a rule
based or machine learning approach to generate and outputs parts of speech in a given
text.
• Word chunking
With using this NLP approach, we can derive usecases from the input business
requirement text. That will mostly be identify noun phrases, and verb phrases and also
propositional phrases using tokenized text and POS tags.

Machine learning would be used to identify meaningful use cases, classes and identify
relationship types in use case and class diagram. And also, we would use this machine learning
techniques to identify multiplicities in the relationship of class diagram. Using machine learning
techniques, we can improve the accuracy level of identifying above mention elements in both
use case diagram and class diagram in a better way even though Rule based method also exists
there as an alternative option.
Fig. 2.1 shows the text classification flow to be considered when applying machine learning
techniques.

C. R. Narawita, K. Vidanage (2017) introduces a technique for generating UML model as use
case and class diagram. This mainly focused on the design phase of a software. In this proposed
system will extract usecase and class diagram elements with including relationships and also
user interaction were involved to do necessary changes. They were used classification model to
identify relationships and they suggested to use regression and they believe that would be a
better approach compared to classification. And also reducing user interaction on this proposed
system would be a good improvisation.

2.1.2 Derive Use case elements from requirement text

This section will briefly review the major research efforts focused on the application of natural
language processing technologies to gain insight from business needs to generate usage and
class diagrams.
Some commercial products that have representing usecase models have been developed such
as visual UML, Rational rose, smart Draw, MS Visio etc. Bajwa,I. S. & Choudhary,M. A.
(2006). Some advanced tools have also been suggested to automate software engineering
activities, which will be more complex than providing advice on drawing possibilities and other
diagrams.
6
Hamza, Z. and Hammad, M., (2019) In this article, he proposes an approach to generate the use
case diagrams from the business requirements using natural language processing techniques.
This proposed approach consists of a few steps to proceed with the requirements. Starting with
the filtering, the text errors and also went through natural language processes all the way to the
generation of the particular use case diagrams.
Osman, M.S., Alabwaini, N.Z., Jaber, T.B. and Alrawashdeh, T. (2019) create a new approach
which will focus on increasing precision of the technique which user requirements convert it to
UML diagrams with reducing time of generating the usecases of requirement text written in
natural language. This will also find solutions to some of the problems in current technologies
as people need a smart and accurate solution to meet their needs with saving time and also the
reliability increasing of the reliance on software.
Elallaoui, M., Nafil, K. and Touahni, R., (2018) present an approach were an Automatic
Transformation of User Stories into UML Use Case Diagrams using NLP Techniques. In there
they propose a process of transforming user stories into use cases and for that they used natural
language processing techniques. But in this solution, it’s not supported for exclude or include
relationships between usecases and also it does not support sentences which will contain more
than one compound noun. The limitations of this paper are derived associations from complex
sentences, which will mostly require exclude or include relationships between usecases and this
type of relationship not solved in this paper. And also, this does not resolve sentences containing
more than one compound noun. The paper also does not mention generalizations and
specializations and usage opportunities between actors and actresses and will not focus on this
solution but in their future work.
Pereira, A., (2018) has proposed a work as using NLP to generate user stories from software
specification in natural language and in there she proposed the user story generation approach
and tool to simplify all the work that eliciting user stories required for software development.
Vemuri, S., Chala, S. and Fathi, M., (2017) Attempts to approach the problem of automation of
the requirements analysis process automation process due to the text format of the requirements.
Use probabilistic technology to identify usecases and actors. This has yielded promising results,
and they state in their paper that this approach, which fully automates the analysis phase, will
be further enhanced.
Further research analysis with research gaps on deriving use case elements to generate use case
diagram can be found in Table 2.1 in 2.2 section.

7
2.1.3 Derive Class Diagram elements from requirement text

This section will briefly review the leading research efforts focusing on the application of
natural language processing to gain knowledge from the business requirement to generate a
class diagram.
Nasiri, S., Rhazali, Y., Lahmer, M. and Chenfour, N. (2020) in this paper they discuss how to
generate a class Diagram presented in the XMI file from the specifications given in the user
stories. Improvements such as quoting the cardinalities of class diagram associations and
applying artificial intelligence to generate new rules, especially for aggregation relationships,
can have better results.
Jaiwai, M. and Sammapun, U. (2017) According to the article, the proposed method is to
process the cable requirements written in Thai to extract UML class diagrams using natural
language processing techniques, and the extraction of UML class diagrams is based on
translation rules for identifying classes and attributes of these classes from the requirement text
provided.
Adhav, V., Ahire, D., Jadhav, A. and Lokhande, D. (2015) in this paper, they used Natural
Language Processing and domain ontology techniques to support class diagram extraction from
human language requirements to find four types of relationships as generalization, association,
composition / aggregation, and dependency. Inhere this Unable to identify one-to-one, one-to-
many, and many-to-many relationships.
Kumar, S.K. (2014) The proposed system would allow developers to create a UML class model
using software specifications using a standard configuration of natural language processing
technologies and class diagrams, and additionally classify the relationship between classes.
More, P. and Phalnikar, R. (2012) This proposed method extracts UML diagrams from text
requirements using natural language processing (NLP) and domain ontology technologies, as
well as finding basic concepts and their connections and extracting UML diagrams.
Further research analysis with research gaps on deriving class diagram elements from
requirement text to generate class diagram can be found in Table 2.2 in 2.2 section.

2.1.4 Generate use case and class diagram

The output of this system is a PlantUML diagram where the project is to generate a use case
and a class diagram using actors, classes, use cases, class attributes, usecase relationships and
class relationships. There are many CASE tools for drawing the UML usecase and class
diagrams. Visual paradigm, StarUML and Rational Rose are some of the famous software tools
currently using in the industry. This project will use an integrated solution for the PlantUML as

8
it is easy to generate language code after deriving the use case elements and class diagram
elements rather than using other tools (PlantUML, n.d.).

2.2 Presentation of Scientific Material

Figure 2.1 ML Model prediction flow

Figure 2.1 showing the Machine learning models prediction flow with using particular ML
algorithm.

Table 2.1 Research gaps on deriving use case diagram elements

Research Topic Deficiencies in Research

Identifying use cases • Identifying meaningful and not meaningful

use cases (Hamza, Z. and Hammad, M., 2019;
Elallaoui, M., Nafil, K. and Touahni, R.,
2018).

Identifying Associations • Identifying include relationships between use

cases from complicated sentences (Osman,
M., Alabwaini, N., Jaber, T. and Alrawashdeh,
T., 2019; Elallaoui, M., Nafil, K. and Touahni,
R., 2018).
• Identifying exclude relationships between use
cases from complicated sentences (Osman,
M., Alabwaini, N., Jaber, T. and Alrawashdeh,
T., 2019; Elallaoui, M., Nafil, K. and Touahni,
R., 2018).
9
Identifying generalization • Identifying actor generalization (Narawita, C
& Vidanage, K. 2016; Elallaoui, M., Nafil, K.
and Touahni, R., 2018).
• Identifying use case generalization
(Narawita, C & Vidanage, K. 2016; Elallaoui,
M., Nafil, K. and Touahni, R., 2018).

Table 2.2 Research gaps on deriving class diagram elements

Research Topic Deficiencies in Research

Identifying multiplicities • In a relationship in class diagram identifying

o one to one
o one to many
o many to many

relationships accurately (Narawita, C &

Vidanage, K. 2016; Adhav, V., Ahire, D.,
Jadhav, A. and Lokhande, D. 2015).

Identifying class attributes and methods • Complete selection of class attributes and
methods by considering the frequency of
words in the document (Nasiri, S., Rhazali,
Y., Lahmer, M. and Chenfour, N. 2020;
Kumar, S.K. 2014).

Identifying Associations • Extract and Identify association classes

from requirements (Nasiri, S., Rhazali, Y.,
Lahmer, M. and Chenfour, N. 2020; Jaiwai,
M. and Sammapun, U. 2017).

Identifying Generalization • Identify class generalization relationships

(Kumar, S.K. (2014; More, P. and Phalnikar,
R. 2012).

Identifying Composition • Identify class composition relationships

(Nasiri, S., Rhazali, Y., Lahmer, M. and
Chenfour, N. 2020; Adhav, V., Ahire, D.,
Jadhav, A. and Lokhande, D. 2015; More, P.
and Phalnikar, R. 2012).

10
 Table 2.3 Existing research evaluation

Research Title What have done Future works/ to do

UML generator – An ✓ Automate UML diagrams • Used classification to
Automated system for from the analyzed identify relationships,
model driven development requirement text using suggested regression
(Narawita, C & Vidanage, NLP. could be a good
K. 2016) ✓ Extract usecase and class approach. (Weka vote
diagram elements algo has used.)
✓ Relationships also • Reduce user interaction
identified
✓ User will interact to do
changes in extracted
elements including
relationship types.
Generating UML Use Case ✓ Provides access to • Future work is to
Models from Software generate UML usage automate the chunking
Requirements Using diagrams from the process with the ability
Natural Language Requirements using to deal with non-
Processing (Hamza, Z. and natural language formatting requirement
Hammad, M., 2019) processing. texts that make NLP
processing easier in the
approach.
• Include, extends,
relationships
Generate use case from the ✓ This technology focuses • Identifying associations
requirements written in a primarily on increasing from complicated
natural language using accuracy and reducing the sentences such as include
machine learning (Osman, time spent on generating or exclude relationships
M., Alabwaini, N., Jaber, T. systems that usecase text between use cases
and Alrawashdeh, T., 2019) written in natural
language.

Automatic Transformation ✓ transform user stories into • include or exclude

of User Stories into UML use cases and take the relationships between
Use Case Diagrams using advantage from all the use cases not yet
NLP Techniques (Elallaoui, work done in the supported.
M., Nafil, K. and Touahni, transformation process of • in their future work
R., 2018) the models according to relationship type, such as
the MDA approach. generalization and
✓ sets of user stories specialization between
generate the UML use actors and use cases will
case diagram be addressed.
automatically. • resolve sentences which
containing more than one
compound noun

A Framework using NLP to ✓ correctly detected actors • Relationships didn’t

automatically convert User- and use cases from user identify.
Stories into Use Cases in stories.
Software Projects (Azzazi,
A. 2017)
11
A Proposed Architecture ✓ This is for assessment of • This will not draw
for Automated Assessment use case diagram where usecase according to the
of Use Case Diagrams teacher will include input but evaluate and
(Vachharajani, V. and problem statement and assess the drawn
Pareek, J. 2014) generate use case diagram diagrams on particular
then for particular format.
problem statement
student’s usecase diagram
will be evaluated.
An Automated Use Case ✓ Read and performing a • This does not have the
Diagrams Generator from full analysis of the user ability to reuse other
Natural Language requirements provided in existing use cases with
Requirements (Zakarya, the English language text. using include, extend and
M., Alqaralleh, B., ✓ It can also generate use generalize relationships
Alemerien, K., Malek, Z., case diagrams.
Alksasbeh and Alramadin,
T. 2017)
An Automated Tool for ✓ Generating the UML • Extending this function
Generating UML Models models like the Usecase to automatically generate
from Natural Language Diagram, Analysis class state chart diagrams for
Requirements model, Collaboration testing class models
(Deeptimahanti, D. and diagram and Design class without the need for code
Babar, M., 2009) model from the natural generation.
language requirements
using more efficient
Natural Language
Processing (NLP) tools.
✓ using available NLP tools
with syntactic
reconstruction rules to
extract required OO
artifacts like actors,
usecases, classes,
operations and attributes.
✓ Identified actors, usecases
and their association
between actors and
usecases.
✓ generate use-case and
class models from Natural
Language requirements
and collaboration and
design class models from
Use-case specifications
along with proper
relationships.
Towards a Generation of ✓ generates a class diagram, • cardinalities of
Class Diagram from User presented in XMI file associations in the class
Stories in Agile Methods format, and from diagram.
(Nasiri, S., Rhazali, Y., specifications which are • test criteria with user
Lahmer, M. and Chenfour, presented in user story stories to build UML
N. 2020) wise.
12
 diagrams dynamically
such as the Activity
diagram.
• Apply artificial
intelligence techniques to
generate new rules,
importantly to detect
aggregation relationships
in between classes.
Extracting UML Class ✓ This has processed the • identifying methods and
Diagrams from requirements written in relationships from
Software Requirements in Thai language to gain requirements to generate
Thai using NLP (Jaiwai, M. UML class diagram using class diagram
and Sammapun, U. 2017) the natural language completely.
processing techniques.
✓ class diagram extraction
based with the
transformation rules
which would identify
classes and it’s attributes
from requirement text.
Generation of UML Class ✓ Semi-automated • Automate the selection
Diagram from Software designers can assist in the of classes, attributes, and
Requirement Specification development of a UML methods by considering
Using Natural Language class model from the frequency of words
Processing (Kumar, S.K. software specifications in the document.
2014) using natural language • Identify different kind of
processing technologies. modules and packages in
✓ class diagram in a the system.
standard setting and
further identifying the
relationship in between
classes.
Generating UML Diagrams ✓ from textual requirements • User interaction
from Natural Language with using the natural involving for add, delete
Specifications (More, P. language processing and renaming classes
and Phalnikar, R. 2012) (NLP) and Domain with its relationships in
Ontology techniques the class diagram.
extracting the UML • Allow the user to view,
diagrams. add, modify, organize
✓ finding basic concepts concepts and
and its relationships, and relationships. The user
extract the UML diagram can easily add new
concepts, change the
concept type and add
new relationships.
Class diagram extraction ✓ Natural Language • Unable to identify one to
from textual requirements Processing and domain one, one to many, and
using Natural language ontology techniques to many to many
processing (NLP) support the extraction of relationships
class diagram from

13
 techniques (Ibrahim, M. Natural language
and Ahmad, R., 2010) requirements.
✓ find four types of
relationships:
Generalization,
Association,
Composition/ aggregation
and dependency.
Semi-automatic Generation ✓ to guide developer in • Multiplicity among
of UML Models from generating UML based classes in class diagram
Natural Language analysis semi-automated is still need to be
Requirements technique has used in addressed in future work.
(Deeptimahanti, D.K. and here and design models • Generation of state
Sanyal, R. (2011) from normalized Natural diagram diagrams from
Language requirements. use cases to test class
✓ capable to visualize UML models without the need
diagrams in any UML to generate code, so that
modeling tool which has the test cases are
XMI import feature init. requirement-based to test
the behavior of the
system.

Table 2.1 and Table 2.2 shows the research gaps on deriving usecase and class diagram
elements. Table 2.3 shows the evaluation of existing research where similar works done.

This chapter mainly focused on finding the best approaches, concepts and techniques to provide
a solution to the problem domain. The chapter started by highlighting the existing researches
done on this domain according to the different techniques in the field of NLP and ML. and also
with the diagram generating techniques. Then presentation of scientific material also provided
in here with research gaps on deriving use case diagram elements and class diagram elements.
In the end detailed existing research evaluation has shown on the table.

14
 CHAPTER 3
PROBLEM ANALYSIS AND METHODOLOGY

In this chapter focused on answering to the questions as how did the research was done and
why it was done in that way. This will cover not only the methods used to collect and analyze
relevant business requirement data, but also the theoretical approach and relevant techniques,
framework used in the term of Natural Language Processing and Machine learning field. This
would inform both the choice of methods and the approach to interpreting those data towards
generating UML use case diagram and class diagram.

3.1 Problem Analysis

Generally, requirements are split into two types as Functional and Non-functional requirements
(Guru99.com. 2019). Requirements that the client or vendor specifically demands as basic
facilities is the Functional requirements and this will be the basic requirements stated by
client/vendor. And also, these functional requirements will help to capture the intended
behavior of the proposed system (Guru99.com. 2019). Non-functional requirements will define
quality attributes of a suggested system and represent various standards that can be used to
identify the operations of a suggested system (Guru99.com. 2019).

Representing the functional requirements in any suggested system use case diagram will be
very helpful and will help in designing a system from the end user perspective (www.aha.io.
n.d.). The use case diagram shows the dynamic aspects of a system with internal and external
interactions. This is an effective technique which will help to communicate the behavior of the
system in the terms of the client by specifying all the behaviors of the system visible from the
outside (Visual-paradigm.com. 2019).

As a design phase, the first step will be to move from the problem domain to the solution
domain. The objective of this phase is to draft the solution of the problem specified in the
gathered requirements. Requirements will indicate as what is needed, design would take it
towards how to satisfy those needs (Computer Notes. 2013). Class diagrams are defined as
static diagrams, and the static view of need is represented by its attributes and functions. They
can also be mapped directly with object-oriented source code (Holland, K. 2018).

15
In the Requirement gathering and design phase will be useful to avoid misunderstanding
regarding the particular software system with involving the users. In order to do this most
requirement engineers, need to analyze requirements and come out with UML Models.
According to the initial investigation carried on during the preparation of this proposal we
identified that, according to the surveys done by various researcher in software industry most
popular UML models are the use case diagrams and class diagrams (Madanayake R. S., 2019).

When drawing these UML Diagrams Requirement engineer or Business analyst need to put lot
of work with gathering and analyzing business requirements and also good amount of time will
waste due to this time taken process. UML Diagrams will draw separately after gathering
requirements and wastage of time will increase when drawing these Diagram using current
diagram tools according to particular business requirement due to complexity of business
scenario or technical capability of UML diagram concepts. Automated UML Diagram
generating tools necessity can be identified with the time spent on analyzing business
requirements and the low quality of the human analysis.

3.2 Proposing Model/design

This study will focus on providing solution to generate Use-Case diagram and Class diagram
against the particular business requirement. system will read and understand the business
requirement using Natural Language Processing and Machine learning and identifying entities
and relationships on that for generate use case diagram. And also, classes and relationship
between classes will also be identified to generate class diagram. Then system will show
generated diagram with including user interacting feature where user can add additional element
and add changes to use case or class diagram or edit existing element and its attribute or
relations. There after user can confirm to generate both use case and class diagram respectively.
The proposed system is limited only in reducing the time taken to draw use-case and class
diagrams against the business requirements in software developments projects.

According to the investigation carried out almost all works that have found was related to
requirement text transform to one particular model diagram either use case or class diagram.
But in this proposed system will review existing NLP and Machine learning techniques which
have been applied to user stories or business requirements and will investigate how well they
meet the requirement of text to model transformation.
16
 Figure 3.1 Proposed Solution Design

As showing in the Figure 3.1 proposed solution design, user can upload a text file containing a
scenario or simply copy and paste the text inside. The system extracts word fragments and
named entities using the NLP module. Subsequently, the ML module will identify and extract
the usecases, actors, classes, associations according to the ML algorithms which have
implemented. Plant UML modeling draws the use case and class diagram. The system allows
editing of the generated diagrams and the user can modify and adjust the use cases, actors,
classes, attributes and all types of relationships. Therefore, the output is a highly customizable
output that the user will be able to achieve the desired output according to the business
requirements provided.

3.3 Methodology

Text data contains a lot of information, but not all of it matters. We can search for names; others
like to get specific links between these named entities. Our goal varies according to our needs.
Sometimes we have to go through all the legal documents to find a legal precedent to validate
your current case or you may need to go through all the research material to find relevant
information on how to cure a disease. There are many other examples such as resume
harvesting, media analysis, email analysis.

17
Imagine when we have to manually go through all the textual data and extract the most relevant
information. Certainly, it is an uphill struggle and you might even end up skipping important
information in the texts.

Therefore, providing an automated way to extract information from text data and present it in a
structured way will help us reap many benefits and significantly reduce the time we have to
spend browsing text documents. This is precisely what information mining strives to achieve.
As the sentences are made up of words that belong to different Parts of Speech which normally
referred as POS. There are eight different Part of Speech in the English language such as noun,
verb, adverb, preposition, conjunction, adjective, intersection and pronoun.

How a specific word functioning in meaning in a given sentence determined by The Parts of
Speech (POS). As an example, take the word “good”. In the sentence, “he was concerned with
establishing and maintaining his good name”, “good” is used as an adjective. Whereas, in the
sentence, “he convinces his father to use his genius for the good of mankind”, “good” is treated
as a noun.

This shows that the Parts of Speech tag of a word is very important when we want to understand
the meaning of a sentence. And this way can influence to extract meaningful information from
our business requirement or functional requirement text.

18
3.4 Algorithmic design

Figure 3.2 Process flow with NLP And ML Model

Figure 3.2 shows the process flow with Natural language processing and Machine learning
model where several preprocessing stages has involved before proceeding to the ML based
information extraction.

Usually, the target information which is related to the requirement text is basically arranged in
a few specific patterns. When we look into the Parts of Speech tagging of a certain sentence,
they follow with a specific pattern. If we can investigate towards those patterns, we can easily
extract the information. Using that extracted information, we can easily create usecase and class
diagram for the project purpose.

For the purposes of algorithmic design, we can choose any pos tagger modules for the English
Literature. Part of the process of marking parts of speech is to mark the words in a story based
on its definition and context for a specific part.

Some POS tagging examples can be mention as: Coordinating Conjunction (CC), Cardinal
Digit (CD), existential (EX), adjective (JJ), modal (MD) such as could, will etc., proper noun
(NNP), predeterminer (PDT), possessive pronoun (PRP), etc. (Rachiele, G.2018) To assign
grammatical information to each word in a sentence we used POS taggers.
19
Most of the information in a requirement text is made up with verb, gerund/present (VBG),
noun plural (NNS), TO, verb base (VB), noun (NN), preposition or subordinating conjunction
with these pos taggers. But they come with different patterns.

Sample requirement text would be like below.

Dataset would be like as Table 3.1 shown below according to the above requirement text.

Table 3.1 Sample dataset

20
For now, according to the given sample, there are 3 kinds of specific patterns that can identify:
• {VBG>NNS>TO>VB>NN}
• {NN>IN>DT>NN}
• {VB>DT>NNS>TO>VB}

Like this way we need to experiment more requirement text for find out more feasible patterns
to figure out the algorithmic features behind this project. Machine learning techniques we can
use to,

• identify actors, usecase and classes

• identify relationship types in use-case diagram and class diagram.

Finding the key terms for class or use case diagram is reflected a classification task in machine
learning. The stories narrate different attributes of a diagram and the task is to identify key
terms for respective attributes. The use case diagram and class diagram have different set of
attributes. The machine learning techniques polarize the data into two clusters; one is training
set and the other is testing test. The features from training instances are used for learning
purpose of the model. The learned model is later on used for prediction of words for unseen
dataset. The comparison of predicted and ground truth key terms assist to estimate performance
of the model.

One of the basic statistical techniques depending on the adjustments among the classification
result by means of probability and overheads that supplement the conclusion. It follows the
theory that problem is probability task and expected values are already available in the space
set. The choice is made with instance with minimal error rate. On contrary, where the cost value
is not being taken into account for decision making, the class with highest probability is chosen.

21
3.4.1 Preprocess Functional requirement texts

All the words in a document have different chances to appear as a key term in respective
attribute. Initially, all the token in a sentence is recognized by means of delimiters. The recurrent
words such as propositions don’t participate as key terms. However, while calculating the term
frequency and inverse document term frequency (TF X IDF), these expressions can dominate
other meaningful expressions. Consequently, to deal with these challenging terms we can
eradicate from script by allocating probability assessment of zero. Additionally, numerals and
non-alphanumeric characters should be removed.

Text pre-processing involves reduction of ambiguities caused by use of several forms of a

certain verb, or singular/plural form of a word. Further, stop words, such as a, the, of, is, do not
carry much information towards our goal of summarization. Described below are multiple
operations employed for preprocessing of documents.

• Document segmentation
A text is divided into several paragraphs to find where each sentence is placed in its
respective paragraph.

• Stemming.
We apply stemming to bring a word to its root or base form. The examples include use of a
singular form rather than using a plural, or removal of ‘ing’ from a verb. To this end,
Stanford NLP stemmer is employed in this paper. Paragraph segmentation. Paragraph
segmentation divides a paragraph into sentences using sentence tags.

• Word normalization.
Each sentence consists of multiple normalized words. Through normalization and
lemmatization individual words become one common form, stemming down to their roots.
Ambiguities are removed by Porter’s algorithm.

• Stop word filtering.

Stop words can be filtered out after performing other preprocessing steps. There is no
uniform rule for selecting a stop word because it depends on individual tasks. In this work,
words such as a, is, in, the, of are selected as stop words and are filtered out from the
document. In text mining applications stop word filtering is considered as a standard step.

22
3.4.2 Machine Learning Techniques to Identify Diagram Elements

3.4.2.1 Classification Model

The training data correlated with computed feature value declared formerly, we can take
advantage of naïve Bayes classifier in order to identify multiple attributes of diagram. The
features for each attribute set are different from other. Hence, we would calculate independent
features for all. Based on this assumption we compute separate features and use naïve Bayes
for extraction of key terms. Using above-mentioned formula provided with training data, a
classification model can be designed and pre-process the data for extracting the key terms such
as actors, use case, class and their relationship. Furthermore, other text features like position
and distance of text can be integrated to improve accuracy. One of the major hypotheses of
naïve Bayes is each feature has independent existence and equally contributes to absolute
outcome.

There are a variety of metrics which can be used to assess how effectively a model works. To
determine the optimal evaluation measure for our model, we must first comprehend what every
metric evaluates. Although we may hear that a trained model is incredibly accurate, based on
the question, what our model is seeking to address, another statistic may be more appropriate
for evaluating the model.

3.4.2.2 Recurrent neural network Unit

Recurrent neural network (RNN) (Jordan, 1997) is being successfully applied to sequential
training issues also including action identification (Donahue et al., 2014), scenery tagging
(Byeon et al., 2015), and speech understanding (Cho et al., 2014). A RNN features a recurring
connectivity, unlike feed-forward systems like convolution neural network (CNN), how the
final hidden stage is an intake towards the forward stage. The following is a description of how
stage is updated,

The intake and concealed states at 𝜏 are time interval are 𝑥𝑡 𝜖𝑅 𝑀 and ℎ𝑡 𝜖𝑅 𝑁 , correspondingly.
The values for the present and recurrent inputs, as well as the bias of the neuron, are 𝑊 ∈

23
𝑅 𝑁𝑋𝑀 , 𝑈 ∈ 𝑅 𝑁𝑋𝑁 , and b 𝜖 𝑅 𝑁 . N is the count of total neurons in this RNN layers, and 𝜎 is a
component wise activation function of the neuron. Due to the repetitive manipulation of the
recurrence weight matrices, RNN learning struggles out from gradient disappearing and
expanding issue. To solve the gradient difficulties, numerous RNN versions have been
suggested, including the long short-term memory (LSTM) (Greff et al., 2017; Jozefowicz,
Zaremba and Sutskever, 2015) and the gated recurrent unit (GRU) (Cho et al., 2014). As a
result, building and learning a broad LSTM or GRU-dependent RNN networks is nearly
impossible. Conventional Convolutional Neural Networks with non-soaked activation
functions like relu, on the other hand, may be layered into a deep neural network and learned
effectively. However, residue interconnections for LSTM systems have been explored in
numerous studies (Pradhan and Longpre, 2016; Wu et al., 2016), no substantial enhancement
has been achieved Furthermore, all extant RNN models have the same ingredient as above
formula.

in which the continuous connections debilitate together all sensory cells. Because the
straightforward display of the outcomes of each sensory cell (Karpathy, Johnson and Fei-Fei,
2015) makes it extremely difficult to determine the functionality of one sensory cell without
understanding everyone else, it is difficult to comprehend and grasp the functions of the
programmed sensory cells (e.g., what sequences each sensory cell reacts to). The autonomously
recurrent neural networks (RNNs) is introduced in this study as a current form of RNN. The
recurring input parameters are, as

3.4.2.3 RNN based Sequence to Sequence Model

This gives a numerous benefit well over standard RNN, which include Machine translation
(Bahdanau, Cho and Bengio, 2014; Klein et al., 2017), headlines creation (Ayana et al., 2017;
Chopra, Auli and Rush, 2016; Rush, Chopra and Weston, 2015), textual summarization
(Chopra, Auli and Rush, 2016), as well as the speech recognition (Donahue et al., 2014) have
all been effectively used using sequence to sequence algorithms (Donahue et al., 2014). In the
(Rush, Chopra and Weston, 2015), influenced with the achievement of intelligent or neural

24
machine translations (NMT), developed the neural consideration the sequence to sequence
design by the based on attention encoder and the Neural Network Languages Model (NNLM)
decoder for extracting key phrase or relation phrase among text, which also outperformed the
conventional technique significantly. Improved strategies were proposed (Sutskever, Vinyals
and Le, 2014) on this approach by substituting a sequence to sequence RNNs with conventional
techniques.

The proposed algorithm includes the recursive encoder and a Recurrent Neural Network and
decoding, and it beats some other best models on widely utilized benchmarked datasets, such
as in (Chopra, Auli and Rush, 2016) has used a rich-feature the encoder to extract buzzwords.
Numerous the text extraction algorithms focused on reducing brief content to phrase
descriptions (Sutskever, Vinyals and Le, 2014). The propose method overcomes numerous
flaws for content extraction in lengthy materials:

✓ They are able to correctly recreate the key elements of primary source materials.

✓ They are able to process long sentences effectively.

✓ They have a tendency not to repeat words conveying same meaning, resulting in
artificial summation.
The proposed recursive RNN based system indirectly blends extracting feature to address the
initial two difficulties.

Such style uses a link to extract words from original sentence retain the context. The real data
can be correctly duplicated. The effectiveness of these Recursive-RNN based mechanisms has
been proven in several following experiments that have attained excellent performance (Jiang
and Bansal, 2018).

3.4.2.4 Strategies of training

Expose the bias and unpredictability of the train and the test measures are two further non-
trivial problems by the present the seq2seq system. The Seq2seq architectures are often learned
with maximizing the probability of ground-truth words provided by their preceding the ground-
truth words and concealed states. Therefore, they are substituted by characters produced by the
algorithm. The interpretation errors can rapidly build throughout sequencing formation since
the produced symbols or words have never been presented to the decoder throughout learning.
Experiential bias is a term used to describe such phenomenon. The discrepancy of readings is
another problem.

25
A curricular training technique called as planned samples is presented to gradually convert the
decoder's feed from the ground-truth words to algorithm produced words. As a result, the
proposed algorithm overcomes the learning and the testing. It is a feasible method of
minimizing susceptibility bias.

3.5 Data acquiring methods

In this research mainly based on business requirement or user stories collections, there will be
a repository of requirement specification documents, text file contains requirements and
documents which contain functional requirements which can be use as training data sets and
also as preliminary test-cases with techniques to process them. Data acquiring methods and
availability of required data for the research are briefly explained in this section.

For the both modules Data will be acquired through online resources and also from various
people and company involve with software developments. Online resources can be various
books, white papers and also freely available repositories in world wide web. From this resource
category I expect to take texts which contains software description, big picture information
about a desired software and specially software requirements or software requirements
specification documents. These texts will explain the idea of a software and what will expect to
find in terms of requirements. In this module expect to have the feature to identify functional
and non-functional requirements most important and valuable resources would be classified
requirement text and software requirement specification documents. Some of the resources that
can be found in world wide web as follow.

• Software requirement dataset can be find in keggle

(https://www.kaggle.com/iamsouvik/software-requirements-dataset).
• Labeled requirement dataset can be find in zenodo (
https://zenodo.org/record/268542#.X05hwMgzZhF).
• Freely available Natural Language Requirements Dataset
(http://fmt.isti.cnr.it/nlreqdataset/).

these NLP and ML models would be evaluated by adding text which containing the
requirements of a software product or project and check result to confirm how far this module
able to identify use-case and class diagram elements through the requirement text. This can be
26
checked easily by referencing the software requirements specification document if available, if
not that can be checked manually. Evaluation method for these models with adding only
functional requirements and check result to confirm whether the output results are correct.
Further we can use, case studies and diagrams drawn for them (as an example BIT Case studies
can be used) and also text books having case studies and answers can be used for the evaluation.

This chapter mainly focused on detailed explanation of problem analysis and methodology of

the proposed system. First in here describing the problem analysis was performed and the

proposed model with design has mentioned thereafter. Then the methodology section started

with explaining the methodology that has been used for the research and algorithmic design

shown at the next section. In there the sample data sets which is prepared for the model
training was shown according to the data set that has been extracted for the project through

various sources as mention in the Data acquiring methods section at last. Mainly in this

chapter preprocessing techniques for requirement texts and machine learning techniques to

identify diagram elements has explained deeply.

27
 CHAPTER 4
IMPLEMENTATION

Having discussed the problem analysis and methodology of the proposed system in the previous
chapter, in this chapter it will be focused on implementation process of the proposed system
with different NLP and ML related techniques and libraries. It starts off with a discussion on
the identifying Use case and Class Diagram related elements with NLP and ML techniques for
the implementation of the prototype followed by detail explanation of the implementation
process and problems encountered with providing proposed solutions upon will also be
discussed in using appropriate code snippets

After preparing the initial dataset as Figure 4.1 we have separate it with each component
according to the diagram element wise. Then we start to working on model building according
to each dataset we have prepared. Python has been used as the programming language and
depending on the model and algorithm we chose, relevant library and framework selected for
the experiment.

4.1 Identifying Use case and Class Diagram Entities with NLP and ML
techniques

After dividing the initial dataset to consists of functional requirement sentences against each
diagram element, we have datasets with actors, usecases and classes separately. Thereafter
each Data set is divided in to following sets:

Having,

• 10 percent validation pairs,

• 80 percent training pairs

• 10 percent test pairs.

To identify each diagram element from the requirement text, separate machine learning model
for each element has built using these training datasets.

In the implementation first preprocessing the data for each model using same the python
function as below. Tokenization, finding unique keywords and removing redundant values,

28
check the words in the english dictionary and removing unused punctuation, removing
stopwords etc. are the preprocessing stages done in the implementation.

Thereafter extracting feature task was perform where it checks existence of whole keyword
and also existence of all parts of known keywords were performed.

29
The recurrent words such as propositions don’t participate as key terms. Calculating the term
frequency and inverse document term frequency (TF X IDF), these expressions can dominate
other meaningful expressions. Consequently, to deal with some challenging terms we can
eradicate from script by allocating probability assessment of zero and also, numerals and non-
alphanumeric characters will be removed.

Thereafter naïve based classifier has been implemented and as previous stages clean and
tokenization tasks also perform before generate predicted tags based on decision rule.

30
In the decision rule method, it takes the functional text, and the keywords from the early stage
perform and also stopwords. First it gets scores of each whole keywords from the functional
text provided. Then get scores for each known keywords with all parts in the requirement
sentence. After that, three steps performed as,

1. adding tag if posterior probability is higher than or equal to 0.5

2. add highest scoring to TF-IDF keywords if not already added.
3. Add ‘c#’ if there no tags (‘c#’ just a special keyword to identify it clearly from other
keywords)

Following above procedures separate models created to identify actors, usecase and classes
from the functional requirement text. Those models trained using training dataset and
evaluation performed using test dataset. We have to perform model training for each actor,
usecase and class Models and store those in a file location to predict using different dataset
(validation dataset) in the evaluation stage.

31
4.2 Identifying Use case and Class Diagram Relationships with NLP and
ML techniques

In this implementation our focus mainly on identifying relationship in both use case and class
diagram related components separately. This is a combination of two models where one model
will be identifying relationship involved entities. As an example of use case diagram first model
will identify what are the actors and use cases involve in the relationship. Second model will
be identifying the relationship type of that particular relationship. These two models need to
build for both use case and class diagram separately. Two models per each diagram type to
identify relationship involved in particular functional requirement text.

As previous implementation, in this implementation also we dividing the initial dataset to

consists of functional requirement sentences against each diagram element which we need this
time usecase relationship and class relationship. Each dataset, again we need to separate for the
two models as explained above paragraph, where one dataset with relationship involved entities
and other dataset with relationship type along with the functional text of each. Each data set is
divided in to following sets:

Having,
• 10 percent validation pairs,
• 80 percent training pairs
• 10 percent test pairs.

To identify each diagram relationship from the requirement text, two separate machine learning
models for each diagram type has built using these training datasets.

Below mentioning the implementation of model which identifying the diagram entities
involving in the Relationship. For this RNN based Sequence to Sequence Model applied train
and build this model. ‘Keras’ framework and its related libraries used for this implementation
and important code snippets shown below from the implementation.

32
In above code input text will transform to a sequence of text and below code will split the
target text.

33
Above function will generate batch with taking input and output text.
After identifying diagram entities involving in the relationship, second model will identify the
relationship type those entities involved with. Important code snippets shown below from that
model implementation.

In below code showing the Convolution Neural network classifier for text classification to
identify relationship type.

34
In below code it adding regularization to output layer and calculate loss and accuracy of each
iteration of input.

35
The proposed model is trained on CPU system with 32GB RAM. In our experiments, dimension
of hidden states and word embeddings are 256 and 128 respectively. For source and target
vocabulary of roughly 22, 00 words is used for training the network. Proposed network is
capable of retaining the context of words; hence we can employ shorter vocabulary size.
Proposed recursive RNN structure introduces very low number of learnable parameters to
network. Recursive property adds 153 and spare parameters in network. Contrary to train
embedding from scratch for smaller number of words we have employed pre-trained embedding
for words representation. We have trained our network with learning rate of 0.0001.
Accumulator value is set to 0.1 using Adagrad. For regularization, we have employed early
stopping. We concurrently run our model in training and evaluation mode. Early stopping
ceases network training when it seems over fit.

4.3 Draw diagram with identified diagram elements

With this implementation we want to generate the usecase and class diagram according to
identified elements through previous two implementation. Usecase and class diagram elements
that has been identified from functional requirement text has outputted as JSON file format.
Sample JSON output structure can be seen as Figure 4.1.

Figure 4.1 Sample JSON output format

36
According to the generated JSON format generate the plantUML markup to render was the goal
of this implementation. After generating plantUML markup text, with using ‘plantuml’ python
library where we can simply generate the respective diagrams. Implementation shown below.

37
This chapter discussed about the implementation process of the proposed UML Diagrams
Generator system. Chapter started with justifying the use of various ML models for the
identifying different diagram elements on the both usecase and class diagram. Mainly in
identifying use case and class diagram entities with NLP and ML techniques describe as the
first section and after identifying use case and class diagram relationships with NLP and ML
techniques described. After wards implementation for the drawing diagram with identified
diagram elements was described. each section in this chapter was describe with the explanation
through code snippets for better understanding of the implementation.

38
 CHAPTER 5
EVALUATION AND RESULTS

In this chapter the results of the implementation are presented and discussed evaluation with
reference to the aim of the study, which was to generating UML diagrams against Business
Requirements Using Natural language processing and Machine learning. The two sub-aims -
the first to find the most appropriate NLP and ML techniques towards identifying use case and
class diagrams elements from business requirement, and the second to generate diagram using
the diagram generating tool form the main comparisons in the evaluation. Evaluation was done
by using a case study and solution given in a text book with the project output.

To describe the functional requirements, we used UML use case diagram and class diagrams as
two different approaches. Natural Language Processing (NLP) and Machine learning (ML)
techniques would use to extract the information because the requirements are written in natural
language. The aim of this project is to setup an approach to automate in generating the UML
use case diagrams and class diagrams from the business requirements text using natural
language processing and Machine learning technologies.

5.1 Collecting Dataset

This proposed system will be adhering to experimental research methodology. Proposed system
will contain with two main approaches to be considered as identify use case elements through
business requirement and identify class Diagram elements through business requirement. Each
approach evaluation data availability briefly explains as follows.

For the both NLP and ML models required Data will be acquired through online resources and
also from various people and company involve with software developments. Online resources
can be various books, white papers and also freely available repositories in world wide web.
From this resource category, expect to take texts which contains software description, big
picture information about a desired software and specially software requirements or
requirement text from software requirements specification documents. These texts will explain
the idea of a software and what will expect to find in terms of requirements. In this module
expect to have the feature to identify use case and class diagram required elements most
important and valuable resources would be classified requirement text and software
requirement specification documents. Some of the resources that can be found in world wide
web as follow.
39
• Software requirement dataset can be found in Kaggle and just needed functional
requirements. (https://www.kaggle.com/iamsouvik/software-requirements-dataset)
• Labeled requirement dataset can be find in Zenodo and need to extract functional
requirements from them. (https://zenodo.org/record/268542#.X05hwMgzZhF)
• Freely available Natural Language Requirements Dataset in National Research council
of Italy with software Requirement specification documents, from them functional
requirements needed to extract. (http://fmt.isti.cnr.it/nlreqdataset/)

40
5.2 Dataset Preparation

Dataset has been created with functional requirements which has been extracted through
above mentioned sources. Sample dataset format model training created as below Figure 5.1.

Figure 5.1 Sample dataset format with functional requirement

Both these NLP and ML models will be evaluated by adding text containing the requirements
of a software product or project and check result to confirm how far this module able to identify
use-case and class diagram elements through the requirement text. Also, it’s worth to evaluate
using a Case study and Solution given in a text book with the model output.

41
5.3 Evaluation approach

Accuracy of identifying elements to draw use case and class diagram can be checked easily by

referencing Case study and Solution given in a text book. Also referring the software

requirements specification document diagrams if those available, if not then that can be checked

by manually identifying. Then we can check result to confirm whether the output results are

same or different. From that we can be able to evaluate that, model was able to identify relevant

elements which are useful to draw complete use case and class diagram.

To evaluate the Model, we can use the term called Model Evaluation Metrics.

Model evaluation metrics are needed to measure model performance. The choice of evaluation

parameters depends on the machine learning task provided. In this project that is classification.

Classification metrics

When performing predictions through classification, four kinds of result that could occur as

follow.

• True positives are when we assume an observation belongs to a particular class and

actually it is belonging to that exact class.

• True negatives are when we assume an observation is not belonging to a particular

class and that actually not belonging to that particular class as expected.

• False positives occur when we predict our observation belongs to a one particular class

but in reality, it is not belonging to that class.

• False negatives occur when we assume an observation is not belonging to a particular

class but actually it is belonging to that same class.

Above mentioned four types of results are usually plotted on a confusion matrix. Following

confusion matrix in Figure 5.2 can be presented as an example for the case of binary

classification.

42
 Figure 5.2 confusion matrix for the case of binary classification

Accuracy, precision, and recall are the three main metrics that can be used to evaluate a
classification model in generally.

Accuracy – In here, this is defining as the percentage of correctly predicted predictions of the
test data. This can be calculated simply with dividing the number of correctly predicted
predictions by the total number of all predictions.

Precision – This would be defined as a fraction of true positives which means relevant
examples among all of the examples as this will predicted to belong in a particular class.
Simply this means how many relevant items were predicted.

43
Recall – This is also defined as the fraction of test data which were predicted as belonging to
a particular class with respect to all of the test data that truly belong in the class. Simply it
means how many relevant items are predicted.

To evaluate both the precision and recall of a model is important. Therefore, it's better to have
one number to evaluate a machine learning model with including this precision and recall
metrics. Thus, it makes sense to have a common approach for combining these metrics. f-
score is introduced to address that concern.

In here β parameter allows to control the tradeoff value as it is important between precision
and recall. When β<1 means focuses more on precision, and β>1 indicating focuses more on
recall. In this case, it’s better to focus more on the model's recall than its precision.

5.4 Evaluation and Results

Detailed discussions of the evaluation results shown below.

5.4.1 Identifying use case diagram element from requirement text

Machine Learning model has able to identified use case diagram element such as actors, use-
cases to acceptable extend since the three main metrics Precision, Recall and F1 score showed
good scores as below.

Identifying Actors
After execution of the program against the functional requirement text identified actors by ML
model and test dataset used to evaluate shown in below Table 5.1, Table 5.2 and Table 5.3.

44
 Table 5.1 Test data snippet for actor in functional text

Table 5.2 Model Predicted results snippet for identified actors

45
The evaluation result of the three main metrics used to evaluate a classification model as
below.

Table 5.3 Evaluation metric result for actors

Identifying Use case

Identifying use cases from the requirement text also had good evaluation metric scores. Test
and it’s predicted data set can find on below Table 5.4 and 5.5.

Table 5.4 Test data snippet for usecase in functional text

46
 Table 5.5 Model Predicted results snippet for identified usecases

Evaluation metrics calculation for identifying usecase from functional requirement text as
below Table 5.6.

Table 5.6 Evaluation metric result for usecases

Identifying use case relationships

Identifying use case relationships was done with two models as explained in Implementation
chapter. First model identifying relationship involved entities. After several training with
parameter changing, it was able to generate good result.
In Table 5.7 shows the data snippet from the preprocessing stage data separation to identify use
case relationship according to involved entity types. As an example, whether that functional
text having actor to use case relationship or use case to use case relationship. Table 5.8 shows
the Test data snippet used to Test the first ML model and Table 5.9 shows the predicted result.

47
Table 5.7 Data snippet from preprocessing stage data separation

48
Table 5.8 Test data snippet for use case relationship involved entities in functional text

49
 Table 5.9 Model Predicted results snippet for identified use case relationship involved entities

Second model will be identifying the relationship type of that particular relationship that can
identify in respective functional requirement text. Table 5.10 shows the Test data snippet used
to test the second ML model and Table 5.11 shows the predicted result of it.

50
 Table 5.10 Test data snippet for use case relationship Type in functional text

Table 5.11 Model Predicted results snippet for identified use case relationship type

51
5.4.2 Identifying class diagram element from requirement text

Trained NLP and ML model able to identify class diagram elements from the functional
requirement input text. This project is only considered on class diagram elements such as
classes and identifying classes that have relationship. Due to limited amount of data relationship
type identifying is difficult with this phase. Identifying classes within the functional
requirement text has acceptable result as the three main metrics Precision, Recall and F1 score
showed good scores as below.
Identifying Classes
Identifying classes from the requirement text had good evaluation metric scores. Test dataset
and it’s predicted data set can find on below Table 5.12 and Table 5.13 respectively.

Table 5.12 Test data snippet for classes in functional text

52
 Table 5.13 Model Predicted results snippet for identified classes

Evaluation metrics calculation for identifying classes from functional requirement text as
below Table 5.14.

Table 5.14 Evaluation metric result for classes

Identifying Class Relationships

Same as usecase relationships, identifying class relationship also done with two models as
explained in Implementation chapter. First model identifying relationship involved entities.
After several training with parameter changing, it was able to generate good result.
In Table 5.15 shows the data snippet from the preprocessing stage data separation to identify
class relationship according to involved entity types. Table 5.16 shows the Test data snippet
used to Test the first ML model and Table 5.17 shows the predicted result.

53
Table 5.15 Data snippet from preprocessing stage class relationship data

54
Table 5.16 Test data snippet for class relationship involved entities in functional text

55
 Table 5.17 Model Predicted results snippet for identified class relationship involved entities

Second model will be identifying the class relationship type of that particular relationship that
can identify in respective functional requirement text. Table 5.18 shows the Test data snippet
used to test the second ML model and Table 5.19 shows the predicted result of it.

56
Table 5.18 Test data snippet for class relationship type in functional text

57
 Table 5.19 Model Predicted results snippet for identified class relationship type

5.4.3 Generating usecase and class Diagram

In here after identifying relevant usecase and class diagram elements from the functional
requirement text using NLP and ML models, particular diagram can be drowned with Plant
UML tool. Generated diagram can evaluate using and comparing the actual diagram that used
to test the program.

Generating Usecase diagram

Generated language code after deriving the use case elements as below in Figure 5.3. In their
identified usecase diagram elements have been put together in a single language code to
generate usecase diagram.

58
 Figure 5.3 Program generated language code for usecase diagram

Syntax error would occur in generated code as Figure 5.3 and user can make changes to the
code with referring identified actor, usecases and their relationship. In below Figure 5.4 shows
the edited version of above Figure 5.3 language code.

Figure 5.4 Edited language code for usecase diagram

59
Figure 5.5 shows the actual diagram that was given in case study and Figure 5.6 shows the
generated usecase diagram according to above Figure 5.4 edited language code.

Figure 5.5 Actual use case diagram according to the functional text

60
 Figure 5.6 Generated usecase diagram using Plant UML tool

Generating Class diagram

Generated language code after deriving the class diagram’s elements as below in Figure 5.7. In
their identified class diagram elements have been put together in a single language code to
generate class diagram.

61
Figure 5.7 Program generated language code for class diagram

Figure 5.8 Edited language code for class diagram

62
After identifying syntax error and incorrect classes and their relationships user can edit the

language code accordingly. Edited language code for above Figure 5.7 would be as above

Figure 5.8

Figure 5.9 shows the actual diagram according to the given functional text and Figure 5.10

shows the generated class diagram according to above Figure 5.8 edited language code.

Figure 5.9 Actual usecase diagram according to the functional text

Figure 5.10 Generated class diagram using Plant UML tool

63
As a conclusion this project prototype has identified some important elements that have
identified by the evaluator, in here it mentioning as actual elements or diagram for particular
functional requirement text. If this project is domain specific most of element would have been
identified the application itself. In general, this proposed diagram generating tool has identified
and generated the use case diagram and class diagram up to a considerable extent which gives
a good idea about the scenario.

This chapter started with describing collecting of dataset and dataset preparation where it
described the data acquiring method from various sources for the both NLP and ML model
training and validation. Then evaluation approach and selected different types of evaluators as
classification metrics to evaluate the different models of the project along with the justifications
for those selections. Evaluation and results were discussed deeply according to the identifying
use case diagram element from requirement text and identifying class diagram element from
requirement text. In there expected results against predicted results shown according to the
snippets of the test dataset and also generated diagram using plant UML tool also be shown
against the actual diagram which is expected according to the test functional text.

64
 CHAPTER 6
CONCLUSION AND FUTURE WORK

The previous chapter presented the results of the evaluation process carried out on the System.

This chapter will be focused on concluding the project by highlighting the achievement of the

goal and objectives, the problems and challenges faced during the life cycle of the project.

And also, in here the limitations of the project, identified future enhancements and closing

remarks describe at the end of this chapter.

6.1 Conclusion

Comparing the research and studies conducted in this context, the generation of UML use case
diagram and class diagram from the natural language text using natural language processing

and machine learning technologies can still be considered a relatively new field. There have

been several research attempts to generate automated use case diagrams and class diagrams

using natural language processing and machine learning methods to extract the element of that

diagram from a functional requirement text or user story text. Also, generating the use case and

the class diagram remain the most challenging areas for finding the ability to analyze and

understand business requirement text as a unit of interpretation to extract important elements.

There was good amount of work related to NLP techniques related to identify UML diagram

elements from functional requirement text or user stories. The work represented by C. R.

Narawita, K. Vidanage (2017) is quite similar to the present one but their approach was more

towards with rule-based approach. According to specific sentence pattern they identifying

usecase and class diagram elements from functional requirement text. They were used

classification model to identify relationships type but that’s not specifying what element would

involve to that relationship. And also, they suggested to use regression and they believe that

would be a better approach compared to classification.

There were some works that automatically generate conceptual models from a series of user
stories in the form of OWL ontology, and our approach to automatically generating UML use

case diagram from functional requirements text or user stories. The advantage of this
65
technology is the ability to reduce the ambiguity of software requirements specifications and to

facilitate the work of the development team and product owner to generate automated design

templates. The advantage of using automate UML use case and class diagram generator is that,

it is easy to understand and helps the developer to interpret the functional requirements or user

stories in one way so that the teams are really integrated in the design process. Also, this allows

designers to save time as they can generate UML use case and Class models from a series of

functional requirement or user stories in a short time.

In this research, the developed prototype can read and fully analyze the functional requirements

provided in English language texts. It can also automatically generate use case diagram and

class diagrams. In our system early stage uses NLP technologies such as tokenization and POS

tagging to interpret system specifications based on a predetermined set of syntactic heuristic

rules. Then, our proposed system includes the ability to analyze and understand input scenarios

with the help of a machine learning model with a classification and also Recurrent neural

network-based Sequence to Sequence Model.

We have taken the advantage of naïve Bayes classifier in order to identify multiple attributes

of diagram. The features for each attribute set are different from other. Hence, we would

calculate independent features for all. Based on this assumption we compute separate features

and use naïve Bayes for extraction of key terms. And also, Recurrent neural network-based

Sequence to Sequence Model used to identify relationships among actors to use cases, use case

to use case and class relationships. Using the training dataset we prepared, a classification

model can be designed and pre-process the data for extracting the key terms such as actors, use

cases, class and their relationship.

In the initial process, the proposed system faced some challenges in research and development

and had to be developed throughout the experiments with different approaches. The author has

done a lot of experimentation and brought it to its present state. Currently the proposed system

has its own ingenuity to extract the essence of UML use case and class diagrams and can

generate use case and class diagrams according to the provided requirement text with
customization. The main objective of this project was to automate generation of use case

diagram and class diagram from the user input text scenario to reduce the time, and cost factors
66
for both system users and business analyst. That input text would be functional requirement text

or user stories in the agile point of view.

Identifying class attributes and multiplicity for classes is very important features in generating

the class diagram. Furthermore, it is important to correctly identify include and extend

relationships in the use case diagram. However, due to the time limit of the research and the

complexity of the task, the author has placed it as a future improvement with the machine

learning model, which requires a high level of intelligence and high accuracy and a good

amount of data for training. The better improved version of a machine learning or deep learning

model should be considered in this regard.

Moreover, the proposed method provides a static and dynamic view for a given business
requirement by generating a class diagram and use case diagram where user can edit the diagram

with Plant UML language code. In relation to the evaluation and testing performed, the system

demonstrates the ability to generate use case and class diagram according to the input text of

the functional requirement or set of user stories. With the use of NLP and ML technologies

researched and implemented by the author, the relevant system will give adequate results in a

reasonable time. But the results can prove that the approach is successful and adaptable to a

variety of business situations. This is simple to prove that the system is pretty much a good

product. The system has the ability to identify business decisions in words, by analyzing the

user's writing style.

In conclusion, the author has done an in-depth analysis and improvements in the field of

generating the use case and class diagram against business requirement. By fine-tuning and

preparing more data for different set of business domains this research may eventually lead to

a commercial product, which will certainly help the system users to get a quick and rough

overview regarding to the system to be developed.

6.2 Future Work

This system can generate use case diagram and class diagram with identifying relationship and
it’s involved component. It does not have the capability to identify relationship types like

include, extend and generalize relationships in use case diagram, and Aggregation
67
Composition, Generalization in class diagram. Therefore, we hope to provide the ability to reuse

existing use cases to reduce the efforts required to define the use cases and class relationship of

the system. Our goal, on the other hand, is to extend our system to automatically identify other

features such as class diagram multiplicity and class attribute. With this phase it couldn’t

achieve as reason of Limited amount of data and Domain variation among data instances.

Improving this sequence-to-sequence learning approach with significant amounts of data may

be targeted for specific domain data in the future.

68
 REFERENCES
✓ Adhav, V., Ahire, D., Jadhav, A. and Lokhande, D. (2015). ‘Class Diagram Extraction
from Textual Requirements Using NLP Techniques’. IOSR Journal of Computer
Engineering, 17(2), pp.27–29.
✓ Ayana, Shen, S.-Q., Lin, Y.-K., Tu, C.-C., Zhao, Y., Liu, Z.-Y. and Sun, M.-S. (2017).
Recent Advances on Neural Headline Generation. Journal of Computer Science and
Technology, 32(4), pp.768–784.
✓ Azzazi, A. (2017). ‘A Framework using NLP to automatically convert User-Stories
into Use Cases in Software Projects’. IJCSNS International Journal of Computer
Science and Network Security, 17(5), p.71. Available.
at:http://paper.ijcsns.org/07_book/201705/20170510. pdf (Accessed: 10 Sep. 2020).
✓ Bahdanau, D., Cho, K. and Bengio, Y. (2014). Neural Machine Translation by Jointly
Learning to Align and Translate. arXiv.org. Available at:
https://arxiv.org/abs/1409.0473.
✓ Bajwa,I. S. & Choudhary,M. A. (2006) “Natural language processing based automated
system for uml diagrams generation,” in The 18th Saudi National Computer Conf.
on computer science (NCC18). Riyadh, Saudi Arabia: The Saudi Computer Society
(SCS), Riyadh, Saudi Arabia, 2006, pp. 1-6.
✓ Barba, P., Lexalytics, (2020). ‘Machine Learning (ML) for Natural Language
Processing (NLP)’. Available at: https://www.lexalytics.com/lexablog/machine-
learning-natural-language-processing (Accessed: 29 Aug 2020).
✓ Bhagat, S., Kapadni, P., Kapadnis, N., Patil, D. and Baheti, M. (2012), ‘Class Diagram
Extraction Using NLP’. International Journal of electronics, Communication & Soft
Computing Science & Engineering, p.1 Available at:
http://www.ijecscse.org/papers/SpecialIssue/comp2/190.pdf (Accessed: 10 Sep. 2020).
✓ Bradbury, J., Merity, S., Xiong, C. and Socher, R. (2016). Quasi-Recurrent Neural
Networks. arXiv:1611.01576 [cs]. Available at: https://arxiv.org/abs/1611.01576
[Accessed 28 Aug. 2021].
✓ Byeon, W., Breuel, T., Raue, F. and Liwicki, M. (2015). Scene Labeling with LSTM
Recurrent Neural Networks. , pp.3547–3555.
✓ Cho, K., Merrienboer, Gulcehre, C., Bahdanau, D., Bougares, F., Schwenk, H. and
Bengio, Y. (2014). Learning Phrase Representations using RNN Encoder-Decoder for
Statistical Machine Translation. arXiv.org. Available at:
https://arxiv.org/abs/1406.1078.
✓ Chopra, S., Auli, M. and Rush, A.M. (2016). Abstractive Sentence Summarization
with Attentive Recurrent Neural Networks. Proceedings of the 2016 Conference of the
North American Chapter of the Association for Computational Linguistics: Human
Language Technologies, pp.93–98.
✓ Computer Notes. (2013). ‘Write a Note on Software Design Phases’. Available at:
https://ecomputernotes.com/software-engineering/write-a-note-on-software-design-
phases (Accessed: 20 Mar. 2021).
✓ Deeptimahanti, D. and Babar, M., (2009). ‘An Automated Tool for Generating UML
Models from Natural Language Requirements’. 2009 IEEE/ACM International
Conference on Automated Software Engineering, pp 680-682.
doi:10.1109/ASE.2009.48.
✓ Deeptimahanti, D.K. and Sanyal, R. (2011). ‘Semi-automatic generation of UML
models from natural language requirements’. Proceedings of the 4th India Software
Engineering Conference on - ISEC ’11, pp. 165-174. doi:10.1145/1953355.1953378.
✓ Donahue, J., Hendricks, L.A., Rohrbach, M., Venugopalan, S., Guadarrama, S.,
Saenko, K. and Darrell, T. (2014). Long-term Recurrent Convolutional Networks for
Visual Recognition and Description.arXiv.org. Available at:
 https://arxiv.org/abs/1411.4389 [Accessed 7 Aug. 2021].
✓ Elallaoui, M., Nafil, K. and Touahni, R., (2018). ‘Automatic Transformation of User
Stories into UML Use Case Diagrams using NLP Techniques’. Procedia Computer
Science, 130, pp.42-49.
✓ Greff, K., Srivastava, R.K., Koutnik, J., Steunebrink, B.R. and Schmidhuber, J. (2017).
LSTM: A Search Space Odyssey. IEEE Transactions on Neural Networks and
Learning Systems, 28(10), pp.2222–2232. Available at:
https://arxiv.org/pdf/1503.04069.pdf.
✓ Guru99.com. (2019). ‘Functional Requirements vs Non Functional Requirements: Key
Differences’. Available at: https://www.guru99.com/functional-vs-non-functional-
requirements.html (Accessed: 1 Sep. 2020).
✓ Hamza, Z. and Hammad, M., (2019). ‘Generating UML Use Case Models from
Software Requirements Using Natural Language Processing’. 2019 8th International
Conference on Modeling Simulation and Applied Optimization (ICMSAO).
✓ He, K., Zhang, X., Ren, S. and Sun, J. (2015). Deep Residual Learning for Image
Recognition. arXiv.org. Available at: https://arxiv.org/abs/1512.03385.
✓ Hochreiter, S. and Schmidhuber, J. (1997). Long Short-Term Memory. Neural
Computation, 9(8), pp.1735–1780.
✓ Holland, K. (2018). ‘What is a Class Diagram?’ Medium. Available at:
https://medium.com/@katie.holland.runs.fast/what-is-a-class-diagram-c74a2129e66c
(Accessed: 20 Mar. 2021).
✓ Ibrahim, M. and Ahmad, R., (2010). ‘Class Diagram Extraction from Textual
Requirements Using Natural Language Processing (NLP) Techniques’. 2010 Second
International Conference on Computer Research and Development, pp. 200-204.
doi:10.1109/ICCRD.2010.71.
✓ Jaiwai, M. and Sammapun, U. (2017). ‘Extracting UML Class Diagrams from
Software Requirements in Thai using NLP’, pp. 1-5. doi:
10.1109/JCSSE.2017.8025938.
✓ Jiang, Y. and Bansal, M. (2018). Closed-Book Training to Improve Summarization
Encoder Memory. arXiv:1809.04585 [cs]. Available at:
https://arxiv.org/abs/1809.04585 [Accessed 28 Aug. 2021].
✓ Jordan, M.I. (1997). Serial Order: A Parallel Distributed Processing Approach. Neural-
Network Models of Cognition - Biobehavioral Foundations, pp.471–495.
✓ Jozefowicz, R., Zaremba, W. and Sutskever, I. (2015). An Empirical Exploration of
Recurrent Network Architectures. proceedings.mlr.press. Available at:
http://proceedings.mlr.press/v37/jozefowicz15.html.
✓ Karpathy, A., Johnson, J. and Fei-Fei, L. (2015). Visualizing and Understanding
Recurrent Networks. arXiv:1506.02078 [cs]. Available at:
https://arxiv.org/abs/1506.02078 [Accessed 8 Aug. 2021].
✓ Klein, G., Kim, Y., Deng, Y., Senellart, J. and Rush, A.M. (2017). OpenNMT: Open-
Source Toolkit for Neural Machine Translation. arXiv.org. Available at:
https://arxiv.org/abs/1701.02810 [Accessed 25 Aug. 2021].
✓ Kumar, S.K. (2014). ‘Generation of UML class diagram from software requirement
specification using natural language processing’. Available at:
https://www.semanticscholar.org/paper/Generation-of-UML-class-diagram-from-
software-using-Kar/cc3acd603c865de6af012f0b1ab803ad86c88402 [Accessed 08 Sep.
2020].
✓ Lei, T., Zhang, Y., Wang, S.I., Dai, H. and Artzi, Y. (2018). Simple Recurrent Units for
Highly Parallelizable Recurrence. arXiv:1709.02755 [cs]. Available at:
https://arxiv.org/abs/1709.02755 [Accessed 28 Aug. 2021].
✓ Madanayake R. S., (2019) ‘Transformation of Requirement Techniques to Reduce
Duplication of Work in Methodologies’, MPhil Thesis,University of Colombo School
II
 of Computing. doi: 10.13140/RG.2.2.11058.22728.
✓ Moldovan, Dan & Surdeanu, Mihai. (2002). ‘On the Role of Information Retrieval and
Information Extraction in Question Answering Systems’. pp. 129-147.
doi:10.1007/978-3-540-45092-4_6.
✓ More, P. and Phalnikar, R. (2012). ‘Generating UML Diagrams from Natural
Language Specifications’. International Journal of Applied Information Systems, 1(8),
pp.19–23. doi: 10.5120/ijais12-450222.
✓ Narawita, C & Vidanage, K. (2016). ‘UML generator - an automated system for model
driven development’. pp. 250-256. doi:10.1109/ICTER.2016.7829928.
✓ Nasiri, S., Rhazali, Y., Lahmer, M. and Chenfour, N. (2020). ‘Towards a Generation of
Class Diagram from User Stories in Agile Methods’. Procedia Computer Science, 170,
pp.831–837. doi: 10.4018/978-1-7998-3661-2.ch008.
✓ Osman, M., Alabwaini, N., Jaber, T. and Alrawashdeh, T., (2019). ‘Generate use case
from the requirements written in a natural language using machine learning’. 2019
IEEE Jordan International Joint Conference on Electrical Engineering and Information
Technology (JEEIT). Available at: https://ieeexplore.ieee.org/document/8717428
(Accessed 5 Sep. 2020).
✓ Pereira, A., 2018. ‘Using NLP to generate user stories from software specification in
natural language’, Universidade Federal do Paraná. Setor de Ciências Exatas.
Programa de Pós-Graduação em Informática, Available at:
https://acervodigital.ufpr.br/handle/1884/58882 (Accessed 6 Sep. 2020).
✓ PlantUML.com. (n.d.). Class Diagram syntax and features. Available at:
https://plantuml.com/class-diagram (Accessed: 10 March 2021).
✓ PlantUML.com. (n.d.). Use case Diagram syntax and features. Available at:
https://plantuml.com/use-case-diagram (Accessed: 10 March 2021).
✓ Pradhan, S. and Longpre, S. (2016). Exploring the Depths of Recurrent Neural
Networks with Stochastic Residual Learning.
✓ Rachiele, G. (2018). ‘Tokenization and Parts of Speech(POS) Tagging in Python’s
NLTK library’. [Medium]. Available at:
https://medium.com/@gianpaul.r/tokenization-and-parts-of-speech-pos-tagging-in-
pythons-nltk-library-2d30f70af13b. (Accessed: 07 May. 2021).
✓ Rush, A.M., Chopra, S. and Weston, J. (2015). A Neural Attention Model for
Abstractive Sentence Summarization. arXiv.org. Available at:
https://arxiv.org/abs/1509.00685 [Accessed 17 Aug. 2021].
✓ Sutskever, I., Vinyals, O. and Le, Q.V. (2014). Sequence to Sequence Learning with
Neural Networks. arXiv.org. Available at: https://arxiv.org/abs/1409.3215.
✓ Vachharajani, V. and Pareek, J. (2014). ‘A Proposed Architecture for Automated
Assessment of Use Case Diagrams’. International Journal of Computer Applications,
108(4), pp.975–8887. doi: 10.5120/18902-0193.
✓ Vemuri, S., Chala, S. and Fathi, M., 2017. ‘Automated use case diagram generation
from textual user requirement documents’. 2017 IEEE 30th Canadian Conference on
Electrical and Computer Engineering (CCECE). doi:10.1109/CCECE.2017.7946792.
✓ Visual-paradigm.com. (2019). ‘What is Use Case Diagram?’ Available at:
https://www.visual-paradigm.com/guide/uml-unified-modeling-language/what-is-use-
case-diagram/ (Accessed: 20 Mar. 2021).
✓ Whitney, E., CODE Magazine (2020). ‘Introduction to Gathering Requirements and
Creating Use Cases’. Available at: https://www.codemag.com/Article/0102061/
Introduction-to-Gathering-Requirements-and-Creating-Use-Cases (Accessed: 25 Aug
2020).
✓ Wu, Y., Schuster, M., Chen, Z., Le, Q.V., Norouzi, M., Macherey, W., Krikun, M.,
Cao, Y., Gao, Q., Macherey, K., Klingner, J., Shah, A., Johnson, M., Liu, X., Kaiser,
Ł., Gouws, S., Kato, Y., Kudo, T., Kazawa, H. and Stevens, K. (2016). Google’s
III
 Neural Machine Translation System: Bridging the Gap between Human and Machine
Translation. arXiv.org. Available at: https://arxiv.org/abs/1609.08144.
✓ www.aha.io. (n.d.). ‘User Stories vs. Requirements’. Available at:
https://www.aha.io/blog/user-stories-vs-product-requirements (Accessed: 1 Sep.
2020).
✓ Zakarya, M., Alqaralleh, B., Alemerien, K., Malek, Z., Alksasbeh and Alramadin, T.
(2017). ‘An Automated Use Case Diagrams Generator From Natural Language
Requirements’, 95(5), pp.1182-1190.

IV
 Appendix A
Following figure illustrate the prototype screenshots with brief description with steps to
followed.

Initial page would be as below where user can upload text file containing the functional
requirement.

V
After uploading the text file user can review the text containing and can edit if needed.

After proceeding with the ‘Extract UML elements’ user can view the identified elements of
each usecase and class diagram elements. And also user able to view the generated language
code for each diagram and see whether generated language code id syntactically correct or
where it has syntax errors if there any.

VI
User can edit each diagram language code with removing or adding elements if there need.
Following figure showing the edited code of above figure which is generated by the
prototype.

After proceeding with ‘Update Diagram’ user will be able to view the usecase and class
diagram generated accordingly.

VII
If there any error with generating diagram that also will show with mentioning the syntax
error with the line of the code as following figure.

User can right click on each image and view each diagram in separate tab to view and
download each diagram in original size. And also the particular URL for generated diagrams
can be seen after opening it with separate tab.

VIII
IX