CHATBOT

Major project report submitted in partial fulfillment of the requirement

for the degree of Bachelor of Technology
in
Information Technology

By

ARPIT SOOD (191513)

UNDER THE SUPERVISION OF

Dr . NISHANT SHARMA

Department of Computer Science & Engineering and Information

Technology
Jaypee University Of Information Technology, Wakhnaghat,
173234, Himachal Pradesh, IND
 Candidate’s Declaration

I hereby declare that the work presented in this report entitled “Android Chatbot ” in
partial fulfilment of the requirements for the award of the degree of Bachelor of
Technology in Computer Science and Engineering/Information Technology submitted in
the department of Computer Science & Engineering and Information Technology, Jaypee
University of Information Technology Waknaghat is an authentic record of my own work
carried out over a period from Feb 2023 to May 2023 under the supervision of Dr. Nishant
Sharma Assistant Professor(Grade-2)Department of Computer Science And Engineering
. I also authenticate that I have carried out the above mentioned project work under the
proficiency stream Data Science.
The matter embodied in the report has not been submitted for the award of any other
degree or diploma.

ARPIT SOOD
Arpit Sood(191513)

This is to certify that the above statement made by the candidate is true to the best of my
knowledge.
(Supervisor Signature)
Dr. Nishant Sharma
Assistant Professor(Grade-2)
Department of Computer Science And Engineering
Dated:

i
PLAGIARISM CERTIFICATE

ii
 ACKNOWLEDGMENT

Firstly, I express my heartiest thanks and gratefulness to almighty God for His divine
blessing makes us possible to complete the project work successfully.

I really grateful and wish my profound my indebtedness to Supervisor Dr NISHANT

SHARMA. Assistant Professor (Grade-2), Department of CSE Jaypee University of
Information Technology. Waknaghat. His endless patience, scholarly guidance, continual
encouragement, constant and energetic supervision, constructive criticism, valuable
advice, reading many inferior drafts and correcting them at all stages have made it
possible to complete this project.

I would like to express my heartiest gratitude to Dr NISHANT SHARMA, Department of

CSE, for his kind help to finish my project.

I would also generously welcome each one of those individuals who have helped me
straight forwardly or in a roundabout way in making this project a win. In this unique
situation, I might want to thank the various staff individuals, both educating and non-
instructing, which have developed their convenient help and facilitated my undertaking

Finally, I must acknowledge with due respect the constant support and patients of my
parents.

Arpit Sood

iii
 TABLE OF CONTENT

Sr. No Title Page No

1 Candidate’s Declaration I

2 Plagiarism Certificate II

3 Acknowledgement III

4 Table of Content IV

6 List of Figures V

7 List of Tables VI

8 Abstract VII

9 Chapter-1 Introduction 1-12

10 Chapter-2 Literature Survey 13-16

11 Chapter-3 System Development 17-47

12 Chapter-4 Performance Analysis 48-62

13 Chapter-5 Conclusions 63-64

14 References 65

iv
List of Figures
Fig No. Title Page No.
1.1 Incremental Model 6

1.2. Gantt chart 12

2.1 System texture 15
3.1 Architectural Design 17
3.2 Architectural Design 18
3.3 Architectural Design 18
3.4 intent invoked 20
3.5 Database Model 21
3.6 UML class diagram 22
3.7 UML activity diagram 24
3.8 UML activity Diagram 25
3.9 UML activity diagram 25
3.10 UML activity diagram 26
3.11 Sequence Diagram 27
3.12 Sequence Diagram 28
3.13 Sequence Diagram 28
3.14 Banking Dialog 32
3.15 Web base Chatbot 33
3.16 Web base chatbot 33
3.17 Scanner for app 34
3.18 Banking bot 34
3.19 API model 43
3.20 API model 43
4.1 User chatbot 60
4.2 Conversion flow 61

v
4.3 Conversion flow 62

4.4 Chatbot Conversion Context 62

vi
 List of Tables

Table No. Title Page No.

3.1 CUI Design Principles 30

3.2 CUI Design Patterns 31
4.1 Simulated User Phrases: Test Results 50
4.2 Simulated User Phrases: Test Results 50

4.3 Simulated Device Testing 53-55

4..4 Overall accuracy comparison of Devices 55

4.5 Conversation exit rate 57

4.6 Subjective & objective measurement 59

vii
 Abstract

Today the users face a lot of problem regarding booking of the hotels in any android
application because in most of the cases the user gets result more than what he expected
or he gets results which are not according to his convenience .This paper focuses on
automating the process of communication by use of chat-bot and it also focuses on
providing customized results to the user which makes the process of hotel booking
convenient and user friendly for him. An extensive research done on existing systems
gave us an insight into their shortcomings which this system attempts to overcome by
creating a chat-bot using Artificial Intelligence Markup Language and using various
algorithms such as Keyword Matching , String Similarity , Spell Checker and Natural
language parser. The implementation of this system has resulted in better resource
utilization and increased responsiveness of user behaviour. This system has been
implemented to integrate with any Hotel Management Android Application to ease the
process of hotel booking.

A chatbot is a conversational agent where a computer program is designed to simulate an

intelligent conversation. It can take user input in many formats like text, voice,
sentiments, etc. For this purpose, many open-source platforms are available. Artificial
Intelligence Markup Language (AIML) is derived from Extensible Markup Language
(XML) which is used to build up a conversational agent (chatbot) artificially. In this paper,
we use „programo‟ which is an AIML interpreter for the generation of the responses of
users’ input. We have used this method for developing an Android application chatbot
that will interact with users using text and voice responses.

viii
 CHAPTER 01

1.1 INTRODUCTION

Once it has a thorough awareness of the user's needs, an AI bot built on

Android can engage with the user. As a result, the AI in the bot will act like
a human and respond to the user's questions rather than one human on one
side and another human on the other. AI Bot makes use of "neural
networks," which are sizable machine networks that resemble the network
of neurons in the human brain. It is a piece of software that use the Internet
to carry out automatic operations. A chatbot typically works more faster
than a human alone can to complete activities that are straightforward and
fundamentally repetitive. An interactive agent, or "bot," is a computer
programme created to mimic an intelligent speech-based communication
with one or more human users.

Chatbots have become increasingly important as a human-computer

interface in recent years. An interpreter, a knowledge base, and a user
interface are the three main parts that make up a chatbot. According to
Laven [6], a chatbot is a computer programme that tries to mimic a written
dialogue with the intention of briefly fooling a human into thinking they are
speaking to someone else. A chatbot is essentially a conversational agent
that can have a natural conversation with the user about a particular topic.
Online chatbots have been widely used for customer support, education,
counselling, and amusement. Famous chatbots now in use include ALICE
[2], Sim Simi, and Clever Bot. Extensible markup language (XML) is the
ancestor of Artificial Intelligence Markup Language (AIML). which is
applied to synthesis a conversational agent. AIML-based chatbots are well-
known for being inexpensive, lightweight, and simple to set up. The AIML
object class of data objects is used to explain how computer programmes
behave. Program-o [1], an open source PHP-based AIML engine, was used
in our article. It is an AIML chatbot script interpreter. Information about the
chatbot is kept in a MySQL database. We also store all AIML scripts in a

1
database. When a user submits a message to a chatbot programme, the
chatbot programme formulates and sends the user a response in
accordance with the AIML response that corresponds to their message.
Under the terms of the GNU General Public Licence, it may be immediately
installed on a local server. Internet-based chatbots can be programmed to
respond to text, speech, and emotional cues. We took user input for this
post in the form of text and voice. Because the user may check the input to
see if there are any problems, text input and output is comparatively
efficient. Text entry, however, takes time. The introduction of a voice
interface with speech recognition technology is the answer. These
techniques greatly enhance the chatbot application's ability to
communicate with the user.

The way people engage with one another and with enterprises has been
completely transformed by digitization and the advent of mobile and
internet-connected devices (Eeuwen, M.V., 2017). As technology companies
integrate artificial intelligence (AI) into the products they sell, such as
Google Assistant, Google Home, and Amazon Alexa, millennials are
adopting and supporting new technologies into the routine of their
everyday lives. Businesses anticipate that the current and forthcoming
generations will be crucial, game-changing clients. They demand smooth
interactions, responses in a matter of seconds rather than minutes, and
more intelligent self-service alternatives. (2017) (Teller Vision). Among the
first businesses to use the technology was the banking and financial
services sector. This integration has significantly expanded and assists
banks in connecting with a larger consumer base, allowing.

1.2 Problem Statement :

Technology called an AI chatbot makes it possible for people and machines

to communicate naturally. We discovered from the literature that chatbots
typically operate like a standard search engine. The essential process flow
remains the same even though the chatbot only provided one output rather
than several outputs or results: a new search will be run whenever an input
is entered. Nothing to do with the earlier output. The goal of this project is

2
to develop a chatbot that can process subsequent searches in light of earlier
ones. This feature will enhance the chatbot's capacity to process input in
the context of a chatbot. Approach: We redesigned the chatbot using a
relational database model strategy in an effort to expand the conventional
chatbot process mechanism.

Users today have numerous difficulties while attempting to book hotels

through any android app since, in the majority of situations, they receive
results that are either unsatisfactory or exceed their expectations. In order
to make the customer's hotel booking experience convenient and user-
friendly, this document focuses on automating the communication process
using a chat-bot and also emphasises giving the user with personalised
results. In-depth analysis of existing systems revealed their flaws, which
this system aims to fix by building a chatbot using an AI markup language
and a variety of algorithms like keyword matching, string similarity, spell
checking, and a natural language parser. The adoption of this technology
improved resource management and increased responsiveness.

1.3 Objectives :

• A chatbot's primary objective is to comprehend the user's needs and

provide the necessary information in response.
• Chatbots enable companies to interact personally with customers
without having to hire human personnel.
• They serve as online assistants who aid with curriculum updates,
paper grading, student and graduate data retrieval, and admissions
process coordination.
• The main objective of the "College Inquiry Chatbot" is to keep users
fully informed about ongoing and upcoming college events, reduce
the workload of the school office staff, save the user's time and
energy, and minimise the time needed to resolve the user's
inquiries.

3
 • Understanding the user's needs and responding with the
appropriate information is a chatbot's main goal.
• Chatbots allow businesses to engage in personal client interaction
without needing to hire human staff.
• They work as online assistants who support curriculum changes,
paper grading, student and graduate data retrieval, and
management of the admissions process.

1.4 Methodology :

To guarantee that a realistic time period is created for each step of the
project and that requirements are properly specified, choosing a suitable
methodology is crucial to the overall development of any software
programme. The creation and design of this programme will take into
account a variety of development approaches. The development
methodology that is most suited for this project will be highlighted in this
section.

1.4.1 Waterfall Methodology :

When you initially learn about software development, you are typically
introduced to this pretty traditional methodology. The requirements
collecting, analysis, design, implementation, and testing phases make up
the five phases of the waterfall model, a highly predictable method for
developing software. The stages are finished one after another. The
fundamental drawback of the waterfall approach is that because the project
is broken up into phases, it is quite rigid. To adhere to the overall project
schedule, each phase has a deadline to provide an output. Project
deliverables, design documentation, and test plans are used to gauge the
success and progress of a project. It is challenging since each project phase
is described early in the project life cycle and goals have been specified.

4
1.4.2 Incremental Model :

The waterfall concept gave rise to this software methodology. Iterative

incremental build phases are used to design, create, and test the
application. Each build will result in the creation of a subsystem or element.
As new requirements are likely to be identified and included in each
incremental release, expanding on the functionality from the previous
build, the project will become more complex overall. Software is frequently
delivered in phases, thus it's crucial to manage the versions of the
software's components throughout its lifecycle using version control
technologies like GitHub. It will only take a few weeks for each build to
complete the app's foundation. It is possible to offer criticism.

Figure 1.1 Incre model

Bugs are found early in short, controllable cycles, making testing and
troubleshooting simpler. As new requirements are easily incorporated into
each build and an updated version is produced, this process is adaptable
during implementation (Khan et al., 2011).

1.4.3 Selected methodology:

The best development methodology for this project is the incremental

model. Since additional requirements are likely to be discovered at later

5
stages of development and since each iterative build makes it simpler to
implement new requirements throughout the development process, the
incremental model's flexibility makes it the best choice for this project.

1.4.4 System Need:

Manually communicating with potential users is a labor-intensive process.

The chatbot and user's one-on-one communication is highly prized.
However, controlling talks between two persons is typically not possible
due to the tens of thousands of applications that are made each year. It
would take a lot of effort for an Android application administrator to track
down suitable solutions and inform each user. Reducing this time-
consuming process is beneficial. By creating an engaging chatbot, the
initiative hopes to lessen the load on the user initiating the conversation
and perhaps other users as well. Both the user and the bot can
communicate via a chat interface.

1.4.5 Existing system:

ELIZA was the first chatbot to be created. Joseph Weizenbaum created it

using the keyword matching method. The concept was to read the user's
input, search for particular patterns, and load the response if the pattern
was discovered. If the keyword is not identified, ELIZA attempts to elicit
further information from the user in accordance with the established rules
in order to maintain contact with them.

Artificial Intelligence Mark-up Language (AIML) files are used by ALICE to

store data and employ pattern matching. Like ELIZA, it is a chatbot whose
primary use is to converse with the user. It is mostly employed for sporadic
user interactions.

• Natasha is an Android hiking companion who responds to your

1.4.6 Problems in the existing system:

6
• ELIZA is speaking in a language she does not comprehend. It only offers
outcomes that follow pre-established rules.

• ALICE can only get information from its database because it lacks the
ability to learn.

Natasha is solely used to speak with users and provide them with
information that is accessible online. The objective of this system is not to
make hotel reservations.

Users have trouble switching between transactions in popular hotel

booking software. In order to complete the transaction, you must navigate
through numerous screens.

1.4.7 The proposed system:

This system interacts with users using a chat application that offers
insightful comments and guidance for obtaining the data required to
reserve hotel rooms. The user receives precise output from the system,
even in the case of small spelling mistakes. Additionally, parsing prevents
transmitting words to the system that don't fit into patterns. Frequent
travellers can maximise the system's potential.

1.4.7.1 GPS tracking:

The Global Positioning System is used by GPS tracking to pinpoint and

follow the precise location of a user's device. The application begins by
following the user's current position, which is then saved in the database.

7
When reserving a hotel room in the same city or when the user forgets to
specify where they want the accommodation, this location is helpful.

1.4.7.2 Calculation of AI:

A chatbot that is programmed in AIML does AI computation. The first step

is for the user to start chatting and provide input. The chatbot is now
determining whether the user's input is unsuitable, inadequate, complete,
or conversational. The user is informed that their contribution was
unsuitable if it is inappropriate. The user is prompted to enter the missing
parameters if the input is insufficient. If the input is conversational, the
chatbot engages the user in a casual conversation. The user receives the
exact output if the input is complete.

1.4.7.3 Hotel reservation:

Users receive personalised recommendations before beginning a hotel

reservation. The user then selects a hotel from the list provided above. The
user is now given a payment link, and when they click it, they are taken to
the payment gateway page. The consumer then pays for the hotel of their
choice. The chat interface is now used to notify the user of the details of
their booking.

1.4.8 Functional requirements:

• Enable users who are not logged in to register and store their data to the
database.

• Send SMS and email confirmation notifications.

• Once the user's credentials have been uploaded to the database, a QR code
using Google's two-factor authentication will be displayed to them, and a
special code will be generated and delivered to the user's mobile device.

• Users must be able to examine information about the accounts they have
access to, such as savings, loans, and checking accounts.

8
The chat bot will send users a transaction statement via email so they may
view their transactions.

• The chatbot's integration of the TrueLayer Starling API returns data on

the user's bank account.

• Users will be able to communicate with the chatbot via voice or text
instructions, and it will be able to understand what they are saying thanks
to Dialogflow API integration's natural language understanding processing.

• When the context from earlier messages and chats may be unclear, a
chatbot should be able to retain a conversational state.

• Offer text- and audio-based answers.

1.4.9 Non-functional requirements:

• The chatbot must be quick to answer, taking no more than 5 seconds to

do so in response to a user's message, for example.

• The database needs to be extensible to handle the increasing number of

users. • The chatbot needs to be dependable and nearly bug- or error-free.

• As sensitive data is used, the chatbot needs to be secure; Google two-

factor authentication will be added as an additional security measure.

• Adhere to data protection regulations as the Privacy Act 1198

• The chatbot may be communicated with using natural language, which

promotes human-computer engagement.

• Give accurate comments and responses for the EEE521 Final Year Project
2017–2018 B00659303 19 input.• If feasible, handle unexpected inputs
effectively and alert the user in the right way.

9
.

1.4.10 Software and Hardware Specifications:

The following is a list of the hardware and software requirements for

implementing a chatbot.

1.4.11 Software specifications:

• Tramp

• Laravel Homestead

• Nginx server

• MySQL DB 11

• PHP 7.1

• JavaScript

• Oracle VM VirtualBox

• Dialogue flow

• Laravel 5.6

• Sublime Text 2

• Ngrok

1.4.12 Hardware Requirements:

A Linux/Ubuntu PC that serves as a local chatbot server Android devices;

Intel Core i5 processor

1.4.13 Project planning:

10
A collection of project milestones and deliverables is developed to show
how the project is broken down so that a management plan can be created
and followed throughout the project life cycle. Minor adjustments could
happen in each iteration, though, because some requirements might need
to be modified. To ensure that all project deliverables are met, the steps
necessary in each iteration are specified. Make a risk management strategy
and project schedule with project tasks for each semester displayed on a
Gantt chart.

1.4.14 Risk management:

A critical step in minimising the influence of hazards on the overall success

of the project is identifying and managing the risks and repercussions
related to any project. Early risk identification enables the implementation
of preventative measures to reduce the possibility of dangers arising. The
following steps will be performed to lower risks in this project: periodically
backing up the project report and critical documents; using Github to back
up code and prevent loss; using your own initiative to keep track of the
project timeline.

1.4.15 Project Management:

Figure 5.4.4.1, a Gantt chart prepared in MS Project, illustrates the project

timeline and shows when milestones should be reached in order to deliver
the project on schedule. The tasks' breakdown, the sequence in which they
must be completed, and the time allotted for each are mentioned below.

11
Figure 1.2 Gantt Chart

12
 CHAPTER 02

2.1 LITERATURE SURVEY

For the advantage of its consumers, the banking sector distributes

technological assets and services through a variety of electronic
distribution channels. According to Ajimon and G.G.S. Gireesh K. (George
and Kumar, 2013), both the financial services and banking sectors consider
online banking to be a business commodity. Many of our services have
entered the digital era thanks to technological advancements, and banking
is one of the main industries that is utilising these advancements to
enhance its offerings. In the UK, there are now two paradigms for online
banking. One of these is an integrated Internet bank with an online
presence but that also uses branches to do business. The second
independent Internet bank, which is totally autonomous and derives all of
its revenue from the Internet alone

Banks use technology to increase their processing speed, attract more

clients, and broaden the range of services they may provide (Consoli,
2005). Over the past ten years, there has been a huge increase in demand
for internet banking. "15% of branch customers use online banking once a
day, 59% once a week, 77% at least once a month, and 53% were confident
that they could do the best part of their banking online" (Barty, J. and
Recketts, T. BBA, 2014). Because consumers can manage their finances
while on the road, online banking has grown in popularity as a way to meet
the demands of modern living and eliminate the necessity for in-person
branch visits. This is clear from the fact that branchless banks are now
emerging .

The majority of internet banking service providers, according to a recent

study by Ling et al. (2016), are attempting to attract many of their clients
to use their services. They view the primary factor as a lack of user
satisfaction with online banking services. Customer satisfaction is

13
primarily influenced by "service quality, web design and content, security,
privacy, speed, and convenience" (Ling et al., 2016). This indicates that
there isn't enough technology to raise client satisfaction. A chatbot might
be integrated into the online banking experience to offer quick, convenient,
and customised services.

The first chatbot to use a pattern matching algorithm is thought to be Eliza.

Joseph Weizenbaum created it in 1964 [2]. A rule-based chatbot built on
the AIML (Artificial Intelligence Markup Language) standard is called
ALICE [3]. It comprises more than 40,000 categories, each of which
combines a pattern with a corresponding reaction. A summary of chatbot
applications created with AIML scripts was presented by Md. Shahriare
Satu and Shamim-AI-Mamun [7]. They claimed that chatbots built using
AIML are simple to use, portable, and effective. Their research offers
comprehensive details on numerous chatbot applications. A chatbot based
on AIML and LSA was created by Thomas N. T. and Amrita Vishwa [4] for
customer support on e-commerce platforms. Their strategy demonstrates
how we might enhance the chatbot's functionality.

2.2 SYSTEM ARCHITECTURE :

14
The following figure gives the information of our system architecture.

Figure 2.1 System Texture

The system operates in both text and voice modes. The first mode is turned
on when the user enters data in text format. The middleware API receives
user input and returns a response. The second mode, on the other hand, is
engaged when the user speaks. To submit the speech data to the
middleware API in this voice mode, we first transform the voice to text.

The mechanism known as middleware is what links AIML scripts to our

Android application. User input is delivered to an algorithm for pattern
matching through AIML scripts when it is received across middleware. A
pattern matching method is used in this procedure to find a match between
a legitimate response and one of the available AIML scripts. The
middleware receives the matching template when the pattern matches.
After that, the middleware converts the template to JSON format and
replies to the Android app. The application decodes the JSON answer after
receiving it and provides it to the user.

15
 CHAPTER 03

SYSTEM DEVELOPMENT

Design

3.1 Architectural Design:

The proposed chatbot's abstract architectural design is shown in Figure in

a simple and understandable manner.

Figure 3.1

16
 Figure 3.2

Figure 3.3

From the online client and the Google Assistant app for Android devices,
users can communicate with the chatbot. Users will communicate with the
chatbot using natural language text or voice phrases. Users will find it
simpler to use and interact with the chatbot thanks to the Google Assistant
17
integration because there will be less typing and effort necessary when
using voice requests and rich responses like graphics and cards. The
chatbot may simply be expanded with Google Home thanks to the
integration of Google Assistant. The client side of the application is created
with Laravel Blade, HTML5, CSS, and JavaScript. The Laravel framework has
a templating engine that adheres to the MVC architectural design pattern.
Blade for Laravel compiles and caches

Blade employs template inheritance, which enables views to dynamically

modify and reuse UI components without changing the code of the view
from which they inherit, as well as derive the fundamental layout of another
view. For control structures that may be necessary in specific view files,
such as the "@foreach" annotation, Blade additionally has its own
annotations (Laravel, 2018). In order for users to readily identify where
they are spending the majority of their money, the client side also makes
use of the chart.js CDN (Content Delivery Service).

Additionally, a web client and Webhook that retrieve JSON data from NLU
via HTTPS POST requests are implemented using the Laravel framework.
As a controller class, Webhook is created, allowing the associated web route
to where the Webhook receives the payload in real time from the NLU.
When the NLU determines that a user has activated a client-side intent,
Google Assistant device, or Home device, the webhook receives data from
the NLU.

An API uses a RESTful design approach and asks data to be obtained, but a
webhook accepts data given to it from another service whenever an update
or event occurs (elastic.io, 2018).

The chatbot can be trained to recognise entities and intents in user speech
using Dialogflow's natural language understanding (NLU) engine. Intent
mapping on the Dialogflow console will be used in the design of Chabot to
direct user utterances to a set of phrases.

18
The diagram shows the data flow as a user invokes an intent: By instructing
Dialogflow to recognise a set of common user expressions, it may
determine whether an intent has been activated. The fulfilment webhook
then receives this data. A custom response is generated and returned to the
user from the webhook either visually or verbally when the required
entities and parameters are extracted along with the action. The webhook
decodes a generated JSON response and sends it back to the user for text
and voice interaction after parsing and validating the JSON from the
received payload.

Figure 3.4 Intent Invoked

3.2 Database design:

Eloquent ORM, which "provides an Active Record implementation for

working with database objects" (Laravel, 2018), is another component of
Laravel. A small scale MySQL Workbench database is used to store the data
locally in order to conserve memory. Each table in a database is represented
by a "model" that enables the definition and manipulation of the
relationships and logic between database items.

Figure 83.2.1 depicts the database model for the proposed chatbot and
defines multiplicity. The users table contains details about application
users, including whether or not they have Google 2 Factor Authentication
enabled. The appointments table contains information on a user's
appointments, including "topic". Contains details provided by the user

19
about the purpose of the meeting. Whether the user already has an
appointment is indicated by the boolean value set for the 'booked' element.
Data pertaining to a user's bank accounts are contained in the Accounts and
Transactions table, for instance, a user may have several different types of
accounts with a single bank, such as checking and savings

Figure 3.5 Database model

3.2 Class Diagram :

Classes required to create a chatbot are depicted in the class diagram in

Figure 9.3.1. Although additional classes may be found or altered
throughout development to satisfy the requirements, this will serve as a
general framework for the chatbot implementation.

20
 Figure 3.6 UML

3.3 Speech Recognition :

By speaking to the chatbot through the built-in microphone of the chatbot

device, users can communicate with the chatbot using voice. Web

21
applications use the HTML5 Speech Recognition API for speech recognition
and communication.

The output voice is then transformed by Dialogflow into JSON objects that
can be read as input. A match score, sometimes referred to as the trust
score, is given to the JSON answer. According to DialogFlow (2015), this
rating demonstrates how successfully the NLU engine matches user input
to the destination indicated in the console. 1 is a real match, and scores
range from 0 to 1. The JSON response is described in more detail below.

"Who are you?" the user says.

3.4 Activity Diagrams :

Activity diagrams are created to outline the processes involved in a chatbot

and to specify the data flow. They depict how consumers behave while

22
dealing with chatbots and how the bots react to those behaviours through
procedures. These will interfere with the chatbot's many features. The logic
used by the chatbot when a user asks a currency conversion is shown in
Figure 10.3.5.1. The user statement is sent to the Fulfilment webhook once
the NLU analyses the user input and establishes the context. The NLU
chatbot's JSON responses are accepted via the webhook, which employs
POST endpoints. The accompanying intended action, which was predefined
in the Dialogflow console during training, is then determined by the
webhook and is mapped to the user-entered statement. If the behaviour

Figure 3.7 UML activity diagram

23
Figure 3.8 UML activity diagram

Figure 3.9 UML activity diagram

24
Figure 1 depicts the data flow when a user asks for their balance. NLU takes
the user's feedback into account while presenting the intended product.
The user's input is subsequently sent to the webhook in JSON format.

Figure 3.10 UML activity diagram

By removing the action from the intent that they receive and determining
if it is contained in the intent, a webhook is utilised to offer users with
appropriate bespoke fulfilment. Depending on the current action, it then
calls the relevant function. The getBalance() function is invoked if the
action is active. This method uses REST endpoints to access the True Layer
API. The NLU chatbot is then delivered the response, which has been
encoded in JSON format. An authentication package is included with the
Laravel framework to handle application authentication. The procedure for
logging in and registering is shown in Figure 123.5.3. Users must register
in the application first in order to create an account. The user will not be
permitted access to the site if they attempt to log in before registering.3.5
Sequence diagrams:

The flowchart in Figure 13 3.6.1 displays a user's behaviour as they ask a

chatbot for their banking information and describes the fundamental steps
taken. The Chatbot NLU (Dialogflow) engine sends user input to the

25
webhook in JSON format. By interpreting the JSON it gets, the webhook
decides what to do with the defined intent published from the NLU. The
webhook will contact the appropriate method in the
TrueLayerClientStarlingHelper class, which will call the True Layer Starling
API and return data to the webhook to generate a response if the action fits
the defined bank procedure. When the webhook receives the information
returned by the API, it encodes the information into JSON, which appears
as a response to the NLU.

Figure 3.11

Figure 3.12

26
 Figure 3.13

The audience that will utilise the chatbot must be carefully considered in
order to implement it successfully. Because mobile banking is important to
a wide range of age groups, the chatbot's target audience is made up of
individuals from various age groups and technological backgrounds. The
chatbot will provide simple, direct responses in order to appeal to all age
groups. Developers can choose a voice for their bot that fits its personality
and use rich answers with Google Assistant (Actions on Google, 2018).
However, as the majority of users of a site like this currently engage online
primarily through messaging services, using a platform like this should be
simple (Interactions.acm.org, 2017).

3.5 Conversational User Interface Design:

The dialogue will take place inside the chatbot using natural language, and
conversational user interfaces (CUIs) will be designed with this interaction
in mind. According to Actions on Google (2018), "Conversation as a method
of interaction is frequently referred to as the new user interface." 2018
(cs.hmc.edu) establish a set of design guidelines for conversational user
interfaces.

27
28
 Table 3.1 CUI Design Principles

A precise list of chatbots, including those utilised in the financial sector, is

recommended by the design process.

29
 Table 3.2 CUI Design Patterns

3.5 Dialog Design

The design of the dialogue is an important factor to take into account when
building a chatbot because the interaction is mostly between the user and
the chatbot and occurs through natural language. To do this, the
appropriate dialogue will be created using follow-up and fallback intents in
the Dialogflow console. There will be a predetermined purpose that, when
used by the chatbot user, will end the conversation in its entirety,
independent of the circumstances. A dialogue box is a group of phrases and
words that react to user input. There will be conversation.

be created with natural-sounding human utterances that respond to users.

Plans for a fallback will be another design feature. The chatbot responds
with a message indicating it does not match the intent input and asks the
user to repeat the question or offer more information if the user statement
cannot be matched to an intent. When communicating with a chatbot,
dialogue is also created via follow-up intents, which promote a more
natural flow of discourse. As an illustration, later intents gather more data
from users as necessary; the user can state;

30
The chatbot utilises successive intents to ask the user for further
information, such as the date, time, and phone number because it is aware
that it needs this information to function properly. As this chatbot is a task-
based interactive chatbot that includes banking domain knowledge,
Dialogflow recommends designing a linear dialogue for interactions where
the data present in the conversation is extracted to help users achieve their
goal. A user statement may be expressed in a variety of ways:

To return the response to the user in this example, the chatbot will remove
the necessary "balance" and "account" inputs (Dialogflow, 2018).

Figure 3.14 Banking Dialog

for Google Assistant, create. Compared to other traditional user interfaces,

it has a more straightforward design that makes it more user-friendly and
effective (DZone, 2018). By pressing the device's microphone or utilising
the input area, users can immediately ask the chatbot a question. When the
icon is clicked, the device's microphone is turned on, the user is made
aware that a recording is being made, and the chatbot's reply is presented
on the screen right away. For each message, Google Assistant employs a
different colour scheme. To distinguish between different chatbot

31
responses, various colours will be employed. and user input.

Figure 3.15 Web base chatbot

the design for the web-based chatbot view. Users can see how

Figure 3.16 Web base chatbot

Users will be able to discover excessive spending by seeing how much they
spend and what they spend their money on in a graphical style. For
instance, they may realise they spend too much on coffee each week. Due
to the fact that each message is given a name and colour depending on
whether it was sent by the user or the bot, it is visually obvious to the user
what the bot's response is.

Users can use an authenticator app on their device to scan the QR code in
the image. Users will be required to enter a special code after each login,

32
which prevents users from seeing the chatbot view until they have verified
their account. After the QR code is scanned, a special 6-digit number will be

Figure 3.17 scanner for app

Figure 3.18 banking bot

Users won't need to switch between several websites to routinely see their
transactions. An email will be sent to the user's registered email address
after they request to examine their transactions.

Implementation:

We're going to create a straightforward app that includes a straightforward

chatbot that will respond to queries we ask of it. See what we will be
creating in this article in the video down below.

33
Step-by-step execution

Step 1: See How to create/start a new project in Android Studio for

information on how to start a new project in Android Studio. Take note that
Java is the programming language you have chosen.

Step 2: Add the dependency listed below to your build.graduated file

Add the following to the build.gradle file by going to Application > Gradle
Scripts. dependency to it in dependencies section.

Go to the AndroidManifest.xml section after adding this dependency to

your project and synchronising it.

Step 3: In the AndroidManifest.xml file, add the internet permission.

The code listed below should be added to AndroidManifest.xml under

Application.

Working with the activity_main.xml file in step 4

Add the following code to activity_main.xml by going to application > res >
layout. The code for the activity_main.xml file is shown below.XML

34
Create a modal class in step 5 to store our messages.

Go to application > java > package name of your programme > right-click
on it > New > Java class and give it the name "MessageModal" before adding
the following code. To make the code more comprehensible, comments are
added.JAVA

35
Step 6: Make a user message layout file.

The drawing folder contains the icons used in this file. Navigate to app > res
> layout, right-click on it, and choose "New > layout source file." Name the
file "user_msg," and then enter the following code there.

Step 7: Design a message layout for the bot.

The drawable folder contains the characters used in this file. To create a
new layout file, select New > Layout File from the context menu of the
application, name it bot_msg, and then enter the following code.

36
Working with the Adapter class in step 8

We must develop an Adapter class in order to set the data for our Chat
RecyclerView objects. Navigate to application > java > package name of
your application > right-click it > New > Java Class, give your class the name

37
MessageRVAdapter, and then add the following code to it. To make the code
more comprehensible, comments are added.

38
39
Create an API key in step 9 to access the chatbot service.

Check out Brainshop.ai With your username and password, you can easily
establish an account. Just register for an account on our website. The screen
below is what you will see after making a new account. You must input your
email address after creating an account and select the Require password
option to request a new password.

After entering your email address, you must update your account's
password. We may now begin writing the API code.

40
 Figure 3.19

To build a new brain for your chatbot, adhere to the aforementioned

instructions. Now that your bot has been created, let's retrieve the API URL
for that brain. You can see the details of your bot as shown below by going
to the settings page in your newly constructed brain.

Figure 3.20

41
Use the MainActivity.java file in step 10.

the code below into the MainActivity.java file. The code for the
MainActivity.java file is shown below.

To make the code easier to read, comments were added.

JAVA:

42
43
44
45
 CHAPTER 04

EXPERIMENTS & RESULT ANALYSIS

4.1 Testing:

Testing is a crucial component of every software development life cycle. In

order to comprehend the software's limitations, this requires carrying out
specific activities and procedures. It is clear that specific problems and
faults are recorded in test cases while testing application restrictions. This
will raise the chatbot's general bar and level of excellence while also
enhancing user experience. To assess the chatbot's overall performance,
several testing techniques were used. The usefulness of the chatbot was
tested through a dialogue, which involved gauging how effectively the
chatbot could comprehend a user-supplied utterance even if you didn't
type it. With average response times for voice and text exchanges, the
response also indicates whether the intent was understood.

4.2 Testing the simulated conversation:

The chatbot's comprehension of user speech for voice and text interactions,
even when it is not written, was examined as part of the dialog's efficacy
evaluation. With average response times for voice and text exchanges, the
response also indicates whether the intent was understood. Performance
data for the chatbot have been tracked through certain testing procedures.

A JavaScript command-line tool called dialogflow was installed in the

project directory specifically for the purpose of running these tests.

46
Metrics were recorded for each simulated encounter, which was utilised to
mimic how users would communicate with the chatbot. The tool
instantiates the Dialogflow API on the command line, fires the welcome
event for it, and then shows the response it returns as the command-line
result to begin the simulated interaction for testing. The command line was
used to test each sort of use case, and JSON formatted results were
produced. Twenty simulated utterances in total were evaluated, and Table
5's findings are displayed there. Each remark tested was specifically word

Table 4.1 Simulated User Phrases: Test Results

47
 Table 4.2 Simulated User Phrases: Test Results

The outcomes of simulated "typical" chatbot conversations with the

command line tool are displayed in the table. The dialogue for the chatbot
was created during the design process using simulated phrases that were
taken from user questionnaires and by watching people interact with the
chatbot. Each statement is given a "Confidence Score" that ranges from 0.0
to 1, with 1 denoting a perfect match and 0 denoting a failure to match the
statement to any intent. However, NLU additionally assigned a full
confidence score of 1 to phrases or utterances where no purpose was
discernible because this correlated with the default fallback intent to
conveniently handle unforeseen user input. It is evident that reaction time
varies depending on the fundamental features a user requires to
accomplish a goal, such as

Only 10% of utterances do not fit the chatbot's intentions, which is a

relatively high percentage. It also accounts for typographical errors. 80%
of the 20 user phrases had successful matching intents overall. These
findings show that the chatbot can comprehend the majority of sentences,
including those with spelling mistakes, which speaks to the conversational
quality. Although the phrase "what I spent this week" (which meaning
"what I spent this week") only received a perfect 1 and a low confidence
score of 0.33, the chatbot was nevertheless able to comprehend the
statement's intent and identify the action, as seen in the figure.

48
typo.

The rounded average response time was 3 seconds, which is a reasonable

amount of time. This implies that users will be able to complete their
required activities through interaction with the chatbot quite fast, which is
also represented in the total length of the dialogue, as shown in Table 76,
which shows the overall interaction time. Instead of using simulated
outcome phrases, the evaluation will employ user test results to further
explain the results.

Through the Actions on Google (AoG) emulator in the AoG interface, Google
Assistant and Home Agent were both evaluated. While Google Home was
tested using voice interaction on the AoG simulator, Google Assistant
integration was tested using text input through the simulator. If the
utterance matched the intent phrase, the simulator delivers a JSON object
with information on the chatbots' level of understanding. Similar language
was used in all facilities to describe any differences in how NLU was
understood.

49
50
51
 Table 4.3 Simulated Device testing

Table 4.4 Overall accuracy comparison of Devices

On numerous simulated devices, extensive testing was done. Google Home

was discovered to be able to discern speech intent even when produced
spoken words after testing on numerous simulated devices. In addition, it
was discovered that Google Home could identify speech in which some
words were missing or mispronounced, as demonstrated in Table 87. It was
discovered that the text input given to Google Assistant on Android phones
had stricter syntax than anticipated. the given input text to match the
defined intent used in the training phrases. This is only true if the user
provides the input text, though, as Google Assistant uses the same speech
recognition and machine learning algorithms regardless of device .
52
Although Google Assistant uses the same speech recognition and machine
learning techniques regardless of the device it is installed into, Assistant
adheres to a tighter NLP syntax when the user provides the input text. Even
if the user's speech is inaccurate, software in Google and Assistant turns it
into the correct input text before it is transmitted to the NLU. Consequently,
Google Home Because the interface was solely voice instructions, Google
Home was able to recognise and detect missing or incorrectly uttered
words. These findings show that voice interaction is the best way to connect
with a chatbot since it can recognise pronunciation problems and respond
appropriately.

As can be seen from the findings in Table 9, the Google Home device was
the primary means of communication with the chatbot since it was able to
accurately match the intent of each delivered speech, earning a score of
100%. When compared to text-based interactions, conversational
interfaces using spoken natural language performed better, with Google
Assistant scoring 57.14%. The two conversational interfaces are very
different from one another, indicating that speaking when interacting with
natural language conversational interfaces is far more effective. With an
average speech recognition rate of 85.71%, Google Assistant's speech
recognition outperforms the HTML5 Speech API, suggesting that Google's
platforms will function more effectively during actual user interactions.

Intent (n) Turns Before Exit

1 3
2 4
3 4
Table 4.5 Conversation exit rate

The table shows how many attempts the chatbot made to understand the
meaning of the misspelt phrase defined in Table 87 7 before terminating
the discussion. The chatbot provided backup responses at every
opportunity, including "Sorry, I didn't quite get that, can you repeat that

53
please?" and "Sorry, I'm having trouble understanding what you said, can
you say it again?" When the chatbot doesn't understand a specific
statement or receives unexpected input, this functionality takes care of it.
Even if the user statement was the identical each time, this also describes
the use of various fallback intentions. Due to the user's frequent requests
for the acceptable phrase and its absence, as well as its inability to, the
chatbot left the conversation.

4. 3 Testing the webhook:

A programme called Postman was used to send HTTPS POST requests to

Webhook in order to verify the custom fulfilment offered by Webhook. This
was done in order to test the Webhook's reaction to a specific user
statement. In a POST request to the Webhook, the request body is retrieved
from the Dialogflow developer console and sent there. The user phrase and
other NLU-returned parameters are represented by a JSON object in the
body, as seen in the picture below.

Bugs can be fixed before the agent is given access to technologies like
Google Assistant and Google Home or web-based chatbots since Postman
is used to verify the operation of the Webhook in the distinct context in
which it was constructed. Each time the response is returned to the event
number, this will identify the network and log the issue. The picture below
displays an example of a JSON response that the webhook returned.

54
To determine whether it could successfully contact the TrueLayer API and
determine any necessary parameters or authentication headers that had to
be given with each subsequent request, the integration of other APIs
implemented inside the project was also put to the test.

4.4 User Testing:

The Google Assistant, home, and web chatbots were chosen to be the three
chatbot communication channels for user testing. As determined by the
user's interaction experience, this will shed light on the chatbot's actual
quality and general usefulness. A list of inquiries was developed to assess
the chatbot during user testing. A group of users were given this
questionnaire, which can be found in the report's appendix, in response to
these inquiries.

4.5 Evaluation:

By outlining performance attributes and examining the outcomes, software

evaluation aims to determine a chatbot's quality. Future work as well as a
reflection of already finished work are desired. The user survey results
shed a lot of light on the general response and accuracy rates. A user group
comprising fifteen people with varied technical backgrounds received
questionnaires. At the School of Engineering and Computing Open Day at
Ulster University, these were filled out by parents and prospective students.
This user base accurately represents the intended market for an app like
this. Results from user surveys and simulated user interactions will be
examined in order to distinguish between subjective and objective
measures. Users were seen to record free-standing data regarding during
the questionnaire.

55
 Subjective (n)Users Objective (n)Users
metrics metrics
Naturalness 44.00% Speech 86.67%80.00%
Recognition
Likeability 93.33% Response 86.67%
accuracy
Ease of use 86.66% Response rate 73.33%
Speech 80.00%
Recognition
Table 4.6 Subjective & objective measurement

Due to low user satisfaction when using their services, as was stated in
Section 1.2, the majority of banks have trouble encouraging their
consumers to adopt the technology. The incorporation of a chatbot into
their financial services would greatly raise the degree of user happiness,
according to the findings in Table 119. This is clear from the fact that
compared to other users, 73.33% of all users believed the chatbot was
"Extremely" fast. considerably quicker than the existing technologies their
bank offers, such an online banking app. The results of the user testing
phase confirmed that chatbots can successfully engage users and boost
satisfaction: 86.66% of all respondents said that the chatbot was either
"extremely very easy" or "moderately easy," easy to talk to, and achieved an
overall likeability of 93.33 out of 93.33%, as indicated in table 119. The
performance and quality of chatbots in actual contact are measured
objectively using the questionnaire results. Overall, the chatbot's voice
recognition rate was quite good; 80.00% of users said the chatbot
understood them "very well." This metric shows the efficacy of simulated
activities used in testing, as seen in Table 7. According to Table 6 of Section
5, the chatbot achieved an overall understanding score of 0.8894 (89.00%)
on the simulated exams. The understanding scale's upper limit is reached
with this outcome.

56
 CHAPTER 05

5.1 CONCLUSIONS

In Section 1.3, Figure 1, the emergence and popularity of chatbots are

obviously depicted. In light of this, the information gleaned from chatbot
testing supports the recent increase in demand from businesses wishing to
implement a chatbot. In comparison to conventional approaches, chatbots
have been discovered to function at a very high level and offer users
trustworthy and prompt replies. The chatbot offers consumers an effective
way to handle their banking, thus users spend relatively little time engaging
with it on average. Due to conversational user interfaces' high degree of
comprehension and speech recognition, users may easily engage with
chatbots to satisfy the demands of modern life thanks to their short
interaction times. The chatbot has demonstrated its capacity to satisfy user
needs for instant access to information and services.

Using sentiment analysis to include user sentiment. NLP methods might be

used to examine a user's transactions to determine their spending patterns.
It would be a smart idea to expand into this market as the Internet of Things
continues to develop, as chatbots generate a significant quantity of data
that is required for the success of IoT and Big Data ideas, which go hand in
hand with further developments in the Internet of Things. Machine
learning and AI. The Google Home line of products might be used to
integrate IoT.

5.2 FUTURE ADVANCEMENT

One of the integration streams originally considered during development

was to integrate the chatbot with Facebook Messenger, but during
development this was revealed by recent media events such as Cambridge
Analytic, a discrediting scandal that resulted in Facebook halting the

57
process of allowing other applications to integrate with its review of
messenger apps, so this integration was not developed because Facebook
stopped allowing developers to create new apps or chatbots through the
service. Although Facebook has just recently reopened its app review
process, it is now allowing developers to re-integrate with Messenger,
allowing development of this integration feature for future prototypes
(Facebook, 2018). Google authentication integration would be a beneficial
feature for an implementation that allows users to link their Google account
across multiple devices where the agent is distributed, such as Google
Assistant, which improves the cross-platform experience for users.

A user can find out financial information about their account, such as how
much they spent in a week. Based on this feature, it was thought that during
the later stages of development, users should be able to directly query the
chatbot to find out where and what they are spending their money on.
Although users can view this information in a graphical format in a web
application, it would also be useful if they could discover it through
interaction with a chatbot.

The chatbot could also be developed to be multilingual. Support for

multiple languages would help eliminate banking in developing countries
and improve the overall availability of technology in the banking industry,
while targeting a larger group of users.

From observing user interaction with the chatbot during user testing, there
were many suggestions to integrate the chatbot into other popular
platforms such as Amazon Echo or Dot, which would increase

58
 REFERENCES

[1] Ling, G., Fern, Y., Boon, L. and Huat, T. (2016). Understanding Customer
Satisfaction of Internet Banking: A Case Study In Malacca. ‘Procedia
Economics and Finance’, 37, pp.80-85.n (Accessed 11/10/2017)

[2] Barty, J. and Recketts, T. (2014). 'Promoting Competition in the UK

banking Industry', Promoting Competition in the UK banking Industry.
BBACompetitionReport.,pp.4.Availableat:https://www.bba.org.uk/wpcon
tent/uploads/2014/06/BBA_Competition_Report_23.06_WEB_2.0.pdf
(Accessed 10/10/2017).

[3] Aburub, F., Odeh, M., & Beeson, I. (2007). ‘Modelling non-functional
requirements of business Processes’. Information and Software Technology,
49(1112),11621171.Availableat:http://dx.doi.org/10.1016/j.infsof.2006.
12.002 (Accessed 10/10/2017).

[4] Ling, G. M., Fern, Y. S., Boon, L. K., & Huat, T. S. (2016). Understanding
customer satisfaction of internet banking: A case study in malacca.
Available at doi:https://doi.org/10.1016/S2212- 5671(16)30096-X

[5] Asif Irshad Khan, A.I., Rizwan Jameel Qurashi, R. J and Usman Ali Khan,
U. A. (2011). 'A Comprehensive Study of Commonly Practiced Heavy and
Light Weight Software Methodologies ', IJCSI International Journal of
Computer Science Issues, 8(4), pp. [Online]. Available at:
https://arxiv.org/ftp/arxiv/papers/1111/1111.3001.pdf(Accessed:08/11
/2017)

[6] Pauline M. McGuirk, P. M. Phillip and O'Neill, P. (2016). 'Using

questionnaires in qualitative human geography', University of Wollongong,
FacultyofSocialSciences,(),pp.11[Online].Availableat:http://ro.uow.edu.au
/cgi/viewcontent.cgi?article=3519&context=sspapers(Accessed:
12/11/2017).

59
60