Web based chat application with webcam using PHP

Seminar II report submitted in partial fulfillment of the

Requirements for the Degree of

Bachelor of Technology
In
Computer Engineering
Submitted by
Mayuri jadhav

DEPARTMENT OF COMPUTER ENGINEERING

S.S.V.P.S.’s B.S. DEORE COLLEGE OF ENGINEERING, DHULE
2023-2024

I
Web based chat application with webcam using PHP

Seminar II report submitted in partial fulfillment of the

Requirements for the Degree of

Bachelor of Technology
In
Computer Engineering
Submitted by
Mayuri jadhav
Guided by
Prof. H. P. Mande

DEPARTMENT OF COMPUTER ENGINEERING

S.S.V.P.S.’s B.S. DEORE COLLEGE OF ENGINEERING, DHULE
2023-2024

II
 S.S.V.P.S.’s B.S. DEORE COLLEGE OF ENGINEERING, DHULE

DEPARTMENT OF COMPUTER ENGINEERING

CERTIFICATE

This is to certify that the Seminar II entitled “web based chat

appluiacation with webcam using PHP” has been carried out by
Mayuri jadhav

under my guidance in partial fulfillment of the degree of Bachelor of

Technology in Computer Engineering of Dr.Babasaheb Ambedkar
Technological University, Lonere (M.S ) during the academic year 2023-
24. To the best of my knowledge and belief this work has not been
submitted elsewhere for the award of any other degree.

Date: Guide
Place: Dhule Prof. H.P.Mande

Seminar co-ordinater
Prof.R.V.Patil

Head Principal
Prof. Dr. B. R. Mandre Dr. Hitendra D. Patil

III
 ACKNOWLEDGEMENT
This Seminar II report has taken its current shape after a lot of hard work and
perseverance-not only just by me. I would like to express our sincere gratitude for
the assistance and support of a number of people who are helping to make this
success.
Immeasurable appreciation and deepest gratitude are extended for the help and
support to Prof. R. V. Patil, my guide for his guidance and enlightening comments
throughout the seminar work. It has been an altogether different experience to work
with him and I would like to thank for his helpful suggestion and numerous
discussions. I gladly take this opportunity to thank Prof. B. R. Mandre (Head Of
Department, Computer Engineering) and Dr. Hitendra D. Patil (Principal, SSVPS,
BSD, College of Engineering, Dhule) for providing facilities during progress of the
thesis.
I wish to express my sincere thanks to Prof.H.P.Mande for his expert, sincere
and valuable guidance and encouragement extended to me.
I am also thankful to all those who helped us directly or indirectly to develop
this thesis and complete it successfully. Then I would like to thank all the Staff for
their encouragement. They had always been very prompt at extending in their
helping hand and sharing valuable technical knows. Special thanks to my Family
and Friends.

Mayuri jadhav

IV
 Table of Contents
Sr.no Page no

List of
figures
1. INTRODUCTION………………………………………… 1

1.1 What are the chat appliacations?.......................................... 1

1.2 The role of PHP in chat appliacation……………………… 2
1.3 Web server environment…………………………………... 2
1.4 Code editor………………………………………………… 3
1.5 Database management system……………………………... 3
1.6 Creating the chat interface…………………………………. 3

2. LITRETURE SURVEY…………………………………………... 4

2.1 Implementation approaches of chat application……………. 4

2.2 Datasets used……………………………………………….. 7
2.3 Evalution…………………………………………………… 8

3. SYSTEM ARCHITECTURE…………………………………….. 10

3.1 System architecture………………………………………… 10

3.2 Chat app or client side……………………………………… 11

4. ADVANTAGES………………………………………………….. 13

5. DISADVANTAGES………………………………………………. 14

6. APPLICATION……………………………………………………. 15

7. CONCLUSION……………………………………………………. 17

8. BIBLOGRAPHY………………………………………………….. 18

V
FIGURE INDEX
Sr.no Name of figures Page
no
Figure 1.1 PHP………………………………………………………………………. 1
Figure3.1 System architecture……………………………………………………… 10
Figure3.2 Package.json file of Client-side…………………………………………. 11
Figure3.3 Client-Side Structure Tree……………………………………………….. 12

VI
 ABSTRACT

Chatting is a method of using technology to bring people and ideas together despite
of the geographical barriers. The technology has been available for years but the
acceptance was quite recent. Our project is an example of a chat server. It is made
up of two applications - the client application, which runs on the user’s web browser
and server application, runs on any hosting servers on the network. To start chatting
client should get connected to server where they can do private and group chat.
Security measures were taken during the last one.
The emergence of new technologies has brought about significant changes in the way
people communicate with each other. One of the most popular ways of communication in
today's digital age is through messaging applications. To facilitate this need, several chat
applications have been developed. In this thesis, we introduce a chat application built
using the MEAN stack, which is a popular technology stack used for building web
applications. This application provides users with the ability to create accounts, join chat
rooms, and send messages to other users in real-time. With the use of web sockets and
Angular's two-way binding, the application allows users to see messages as soon as they
are sent. Moreover, the application also includes features such as user authentication
and authorization to ensure secure access to the chat rooms. Through this project, we
aim to demonstrate the feasibility and effectiveness of building real-time chat
applications using the Mern stack.

VII
 Chapter - 1
INTRODUCTION

This chapter gives Simply put, your dissertation’s introduction chapter needs to provide an overview of
your planned research, as well as a clear rationale for it. In other words, this chapter has to explain the
“what” and the “why” of your research – what’s it all about and why’s that important.

1.1What are Chat Applications?

Chat applications, often referred to as instant messaging apps, enable individuals and groups to
exchange text, multimedia, and even conduct voice and video calls in real-time over the internet. These
applications have transformed the way we communicate, making it possible to connect with people
across the globe instantly.

Chat applications are used in various domains, including:

• Social Networking: Platforms like Facebook Messenger, WhatsApp, and Snapchat facilitate
personal and group conversations.
• Business Collaboration: Tools like Slack and Microsoft Teams help teams collaborate and
communicate effectively.
• Customer Support: Live chat support on websites allows businesses to assist customers in
real-time.
• Gaming: In-game chat features enable players to strategize and interact while gaming.
• Education: Chat applications are used in online learning platforms for student-teacher
interactions.

Figure 1.1 PHP

1
1.2 The Role of PHP in Chat App Development
PHP (Hypertext Preprocessor) is a versatile and widely-used server-side scripting language. It is
renowned for its ability to handle web-based tasks and server-side operations efficiently. When it
comes to developing chat applications, PHP offers several advantages:

1. Real-time Communication: PHP can be used to implement real-time features in chat

applications. It can handle server push mechanisms like WebSockets, making it suitable for
instant messaging.
2. Compatibility: PHP is compatible with various databases, making it easy to store and
retrieve chat messages and user data.
3. Rich Ecosystem: PHP has a vast community and a rich ecosystem of libraries and
frameworks that simplify the development process.
4. Cost-effective: PHP is open-source, making it a cost-effective choice for startups and small
businesses looking to develop chat applications without high licensing fees.
5. Scalability: With proper architecture and design, PHP-based chat applications can be scaled
to accommodate a growing user base.
Choosing the right technology stack for your chat application is crucial. PHP stands out as an excellent
choice for several reasons:

• Ease of Learning: PHP has a relatively low learning curve, making it accessible to
developers with varying levels of experience.
• Community Support: PHP has a vibrant and active community that continuously
contributes to its development. This means you can find ample resources and support online.
• Rapid Development: PHP’s simplicity and versatility allow for quick development and
deployment of chat applications, helping you get your product to market faster.
• Cross-Platform Compatibility: PHP works seamlessly across different operating systems
and web servers, ensuring broad compatibility.
1.0.3Required Tools and Technologies

Before we get started with building our PHP-based chat application, let’s ensure that you have the
necessary tools and technologies in place. Here’s what you’ll need:

1.3Web Server Environment

To run PHP scripts locally for development, you can set up a web server environment. You can choose
from various options, but some of the most common ones include:

• XAMPP: XAMPP is a free and open-source platform that bundles Apache, MySQL, PHP,
and other useful tools. It’s available for Windows, macOS, and Linux.
• MAMP: MAMP is another popular choice for macOS users, offering a simple way to set up
a local server environment.
• WAMP: If you’re using Windows, WAMP provides an easy-to-install package with
Apache, MySQL, and PHP.
• LAMP: For Linux users, LAMP (Linux, Apache, MySQL, PHP) is a classic stack that can
be configured manually.

2
1.4Code Editor
You’ll need a code editor for writing PHP code. Some popular choices include:

• Visual Studio Code: A lightweight, free code editor with excellent PHP support and a wide
range of extensions.
• PHP Storm: A robust PHP-specific IDE with features tailored for PHP development.
• Sublime Text: A highly customizable and lightweight code editor suitable for PHP
development.
1.5 Database Management System
For storing chat messages and user data, you’ll need a database management system. MySQL and
PostgreSQL are commonly used choices. Make sure you have the database system of your choice
installed and configured.

1.3.1 Setting Up the Development Environment

Now, let’s walk through the steps of setting up a basic development environment using XAMPP as an
example:

1. Install XAMPP:
• Download and install XAMPP from the official website based on your operating
system.
2. Start Apache and MySQL:
• Launch the XAMPP Control Panel and start the Apache and MySQL services
3. Create a Project Folder:
• Create a new folder in the “htdocs” directory (usually located in the XAMPP
installation folder) to host your chat application project.
4. Write Your First PHP Script:
• Open your chosen code editor and create a new PHP file within your project
folder.For example,name it“index.php.”
<?php
echo“Hello,World!”;
?>

1.6 Creating the Chat Interface

A user-friendly and visually appealing chat interface is crucial for a successful chat application. It
enhances the user experience and encourages engagement. In this section, we’ll explore how to create a
basic chat user interface using HTML, CSS, and PHP.

3
 Chapter - 2
LITERATURE SURVEY
In this section we outline the following main aspects of chatbots based on our finding from the
literature review: implementation approaches, available public database used in previous data-driven
approaches to chatbot implementation, the main evaluation methods for measuring the performance of
chatbots and the application of chatbots in different domains.

2.1. Implementation Approaches to Chat application

In this section, we will give an overview of chatbots’ implementation methods. We will distinguish
between Rule-based chatbots, and Artificial Intelligence (AI) based chatbots. Within AI-based chatbots,
we will further distinguish among Information-Retrieval chatbots and Generative Chatbots. We will also
discuss drawbacks and limitations of each implementation approach, as well as recent improvements.
The very first attempts at chatbots’ implementation were rule-based. Rule-based models are usually
easier to design and to implement, but are limited in terms of capabilities, since they have difficulties
answering complex queries. Rule-based chatbots answer users’ queries by looking for patterns matches;
hence, they are likely to produce inaccurate answers when they come across a sentence that does not
contain any known pattern. Furthermore, manually encoding pattern matching rules can be difficult and
time consuming.Furthermore, pattern matching rules are brittle, highly domain specific, and do not
transfer well from one problem to the other.
AI models, contrary to Rule-based models, are based on Machine Learning algorithms that allow
them to learn from an existing database of human conversations. In order to do so, they need to be
trained through Machine Learning algorithms that can train the model using a training dataset. Through
the use of Machine Learning algorithms, there is no longer the need to manually define and code new
pattern matching rules, which allows chatbots to be more flexible and no longer dependent on domain
specific knowledge. As stated, AI models can be further categorised into Information Retrieval based
models and Generative models.
Information Retrieval Models. Information Retrieval based models are designed so that given a
dataset of textual information, the algorithm will be capable of retrieving the information needed based
on the user’s input. The algorithm used is usually a Shallow Learning algorithm; nonetheless, there are
also cases of Information Retrieval models that use Rule-based algorithms and Deep Learning ones.
Information Retrieval based models include a pre-defined set of possible answers; the chatbot processes
the user query and based on this input it picks one of the answers available in its set. The knowledge
base for this kind of model is usually formed by a database of question-answer pairs. A chat index is
constructed from this database, in order to list all the possible answers based on the message that
prompted them. When the user provides the chatbot with an input, the chatbot treats that input as a
query, and an Information Retrieval model akin to those used for web queries is used to match the user’s
input to similar ones in the chat index. The output returned to the user is thus the answer paired with the
selected question among those present in the chat index The main advantage of this model is that it
ensures the quality of the responses since they are not automatically generated. These models have seen

4
a surge in popularity with the advent of the Web 2.0 and the increase in available textual information
that could be retrieved on social media platforms, forums, and chats
One of the main downsides of this approach is that creating the necessary knowledge base can be
costly, time-consuming, and tedious. Furthermore, if the great volume of data available provides for a
greater training set and aider knowledge base, it also implies it will be all the more challenging to match
a user’s input to the correct answer; a significant amount of time and resources must be deployed to train
the system to select one of the correct answers available
Finally, Information Retrieval systems, due to the fact that they do not generate answers but rather
retrieve answers from a pre-defined set in their knowledge base, are arguably less suitable to be used as
the underlying algorithm for conversational or chit-chat agents-the so-called social chatbots. Information
Retrieval models are in fact less suitable to develop a personality, which is an important trait for this
kind of chatbot Nonetheless, some progress has been made in developing new Information Retrieval
algorithms in recent time, and it is worth mentioning what Machine Learning algorithms are currently
being used as underlying technology for this kind of model. proposed a new model to represent local
textual co-occurrence and map hierarchical information across domains for more semantically distant
terms. This model was based on the idea that the higher the co-occurrence of two terms across domains,
the more closely related the two terms are. Accordingly, a high co-occurrence within a specific domain
could inform the information retrieval process. This model was, thus, based on two steps: topic
modelling for parallel text, and getting the hierarchy architecture. The first step aims at finding
meaningful co-occurrence patterns of words. The second step aims at modelling the architecture of co-
occurrences across topics. This architecture will be used to create the neural network that powers this
machine learning algorithm. The interesting development made by this model lies therefore in its use of
co-occurrences of words to define a context. The underlying aim of this research was to use contextual
information to improve matching performances for Information Retrieval models
One interesting development, which aims at taking into consideration previous turn in the
conversation, thus obtaining more contextual information in order to improve the quality and the
correctness of the output is the one proposed by In this model the Information Retrieval process is
enhanced by a Deep Neural Network that ranks not only the question/answer pair matched with the last
user’s input, but also those question/answer pairs that match with reformulated versions of previous
conversation turns. The ranking lists corresponding to different reformulations are then merged. In this
way, contextual information can be leveraged from the user’s previous queries, and these pieces of
information can be used to retrieve a better answer within the knowledge base
Generative Models. Generative based models, as the name suggests, generate new responses word
by word, based on the input of the user. These models are thus able to create entirely new sentences to
respond to users’ queries; however, they need to be trained in order to learn sentence structure and
syntax, and the outputs can somewhat lack in quality or consistency
Generative models are usually trained on a large dataset of natural phrases issued from a
conversation. The model learns sentence structure, syntax, and vocabulary through the data that it has
been fed. The overall aim is for the algorithm to be able to generate an appropriate, linguistically correct
response based on the input sentence. This approach is usually based on a Deep Learning Algorithm
composed of an Encoder-Decoder Neural Network model with Long-Short-Term-Memory mechanisms
to counterbalance the vanishing gradient effect present in vanilla Recurrent Neural Networks

5
 Industry-Standard Algorithms. Among AI models, Sequence to Sequence models have become the
industry standard for chatbot modelling. They were first introduced to solve Machine Translation
problems, but the underlying principles do in fact seem to perform well for Natural Language
Generation as well. These models are composed of two Recurrent Neural Networks (RNN), an Encoder
and a Decoder. The input sentence of the chatbot user becomes the input of the Encoder, which
processes one word at a time in a specific hidden state of the RNN. The final state
This model does offer some interesting advantages. First, it does not involve domain-specific
knowledge, but is rather an end-to-end solution that can be trained using different datasets, thus on
different domains. Furthermore, although the model does not need domain-specific knowledge to
provide valuable results, it can be adapted to work with other algorithms if further analysis on domain-
specific knowledge is needed. It is thus a simple yet widely general and flexible model that can be used
to solve different NLP tasks For these reasons, the Sequence-to-Sequence model seems to have become
the industry standard choice for dialogue generation and many NLP tasks in recent years. Nonetheless, it
has a considerable limit: the entirety of the information contained in the input sentence must be encoded
in a fixed length vector, the context vector, and thus, the longer the sentence, the more information gets
lost in the process. That is why Sequence to Sequence models do not perform well when they must
respond to longer sentences and tend to give vague answers. Furthermore, when generating an answer,
these models tend to focus on a single response, which creates a lack of coherence in the turns of a
conversation
Transformers. Of course, one of the most interesting innovations in Deep Learning language
models has been the introduction of Transformers, first presented in the paper “Attention is all you
need”. Transformers are language models based solely on the Attention mechanism. Transformers are
nowadays the model of choice for NLP challenges, replacing RNN models like long short-term memory
(LSTM) by differentially weighing the relevance of each portion of the input data. Furthermore, they
provide training parallelization that permits training on larger datasets than was originally achievable.
This led to the development of pretrained systems such as BERT (Bidirectional Encoder Representations
from transformersand GPT (Generative Pre-trained Transformer), which were trained with huge
language datasets, such as Wikipedia Corpus and Common Crawl, and may be fine-tuned for specific
applications. Several different versions of the Transformer have since been presented, such as the
Reformer and the Transformer XL [Each version of the transformer has been developed to answer to
specific challenges for the task at hand. Even though transformers were introduced to answer Machine
Translation challenges, they can be adapted and modified to perform dialogue Modelling tasks.
In the authors propose an updated version of the Transformer called Transformer-XL. This model
can go beyond the fixed-length context limitations of the Transformer, using sentence-level recurrence.
Transformers show a potential of learning longer-term dependency but are constrained by fixed length
context in the setting of language modelling. The authors present a unique neural architecture named
Transformer-XL that enables learning dependency beyond a given length without breaking temporal
coherence. It comprises a segment level recurrence mechanism and a unique positional encoding
technique. This solution aims at capturing longer-term dependency and resolving the context
fragmentation issue. Even though this approach has not yet been applied to dialogue modelling, it can be
argued that once the appropriate and necessary modification implemented, it could prove useful in
overcoming some of the issues current dialogue models present, namely context understanding.

6
 In the authors introduce the Reformer, a more efficient version of the Transformer, that makes use
of two techniques to improve the Transformer in terms of efficiency. Firstly, the authors substitute dot-
product attention with one that employs locality-sensitive hashing, increasing its complexity
from 𝑂(𝐿2)𝑡𝑜𝑂(𝐿𝑙𝑜𝑔𝐿)𝑂(𝐿2)𝑡𝑜𝑂(𝐿𝑙𝑜𝑔𝐿), where L is the length of the sequence. Secondly, they employ
reversible residual layers instead of the standard residuals, which permits storing activation only once in
the training process instead of N times, where N is the number of layers. As a result, the Reformer is
significantly more memory-efficient and substantially faster on longer sequences.
In the authors introduce Meena, a generative chatbot model that was trained end-to-end on 40
billion words extracted and filtered from public domain social media discussions. With Meena, the
authors stretch the limits of the end-to-end approach in order to show that a big scale low-perplexity
model can produce quality language outputs. The authors employ a seq2seq mode
with the Evolved Transformer [as the main architecture. The four most important characteristics of
the Evolved Transformer’s architecture are the utilization of (i) large depth-wise separable convolutions,
(ii) Gated Linear Units, (iii) branching structures and (iv) swish activations . Both the Evolved
Transformer’s encoder and decoder independently generated a branched lower portion with wide
convolutions. Also in both situations, the later portion is essentially identical to the Transformer. The
model is trained on multi-turn dialogues where the input sentence is comprised of all turns of the context
(up to 7) and the output sentence is the response. To quantify the quality of Meena and compare it with
other chatbots, also propose an effective human evaluation metric. Sensibleness and Specificity Average
(SSA) incorporates two basic qualities of a human-like chatbot: making sense and being specific in their
response. The authors ask human judges to evaluate responses based on these two aspects.

2.2. Datasets Used

In this section we will discuss the datasets that appear to be most frequently used to train deep
learning chatbot models. Firstly, we must differentiate between open domain datasets, and closed-
domain datasets. Within the literature, there appear to be a few open domain datasets that are most
commonly used:
OpenSubtitles, Cornell, and the DailyDialog dataset. We will discuss each one of these datasets
more in detail. On the other hand, there appear to be no closed domain dataset repeatedly used in the
literature surveyed. This can be due to the specificity of such datasets, that usually attend to specific
needs and objectives within a precise scope. For brevity, we will not describe each one of these datasets
in details.
OpenSubtitles. The OpenSubtitles database is used by several works of literature. It is an open
domain database made of movies’ subtitles in XML format. The database offers open access movies
subtitles for several languages. The dataset appears to be quite large, containing millions of sentence
pairs, however the data does not appear to be of the best quality; scene descriptions, close captioning,
and segmented sentences appear among the dialogues, which can be problematic when training an open
domain chatbot, since the cohesiveness of the dialogue can be lost
Cornell. The Cornell Movie-dialogue is also quite used to train open domain dialogue system. This
corpus presents a metadata-rich collection of fictional conversations taken from raw movie screenplays,
consisting of more than 300,000 total utterances. The data is presented in ‘txt’ format. Since the data has
been taken from movie scripts, it appears to be of decent quality. However, the dataset might not be
large enough to train more advanced language models
7
DailyDialog. This dataset is human-made and consist of several multi-turn conversations. There are little
over 13,000 conversations, each composed of 8 turns per speaker on average. The data is annotated:
annotations contain information regarding emotions and intents. These pieces of information can be
useful in training some language models. Although the data appears to be clean and less noisy compared
to other datasets, it is also smaller, therefore less suited for more complex dialogue models

2.3. Evaluation
Evaluating dialogue systems has proven to be a challenging task, since human conversation attends
to different goals and functions. Depending on the aim of the chatbot, the metrics used to evaluate the
dialogue can change. A personal assistant chatbot will be mostly evaluated based on the effectiveness of
the interaction (did the chatbot complete the task the user asked? Was the exchange efficient?), whereas
a companion chatbot will be evaluated on its ability to keep the conversation going and to engage users.
There are two main ways to evaluate a chatbot: human evaluation and automated evaluation metrics.
Human evaluation consists of asking a group of participants to interact with the chatbot, and then
evaluate the different aspects of the interaction according to an evaluation frameworks or questionnaire.
Participants will usually rate the different aspects of the interaction based on a scale that can be used to
draw averages and measure the quality of the performance in terms of efficiency, effectiveness, and
users’ satisfaction /Although human evaluation allows to assess the quality of different aspects of the
interaction, it is costly (since there is a need to allocate human resources for the evaluation), time
consuming, not easily scalable, and subject to bias (even when following an evaluation framework,
different people can rate the same interaction differently). Nonetheless, human evaluation can take into
consideration several aspects of the exchange and assess the conversation at different levels; moreover,
the evaluation framework can be adapted based on the main aim and functions of the chatbot or dialogue
system. For these reasons, human evaluation metrics are used in several pieces of literature analysed,
such as The PARAdigm for DIalogue System Evaluation is one of the most extensively utilized
frameworks for combining different levels of evaluation (PARADISE). Firstly, PARADISE evaluates
subjective variables such as (i) system usability, (ii) clarity, (iii) naturalness, (iv) friendliness, (v)
robustness to misunderstandings, and (vi) willingness to use the system again. It accomplishes this by
soliciting user feedback via the dissemination of questionnaires. Secondly, through optimizing task
success and decreasing dialogue costs, PARADISE aims to objectively quantify bot efficacy propose
different frameworks for human evaluation of chatbots. However, since these frameworks are reliant on
human evaluation and therefore not easily scalable The authors argue that trust is at the heart of effective
human-chatbot interaction and examine how trust as a meaningful category is being redefined with the
introduction of deep learning-enabled chatbots. However, the proposed evaluation metric does not seem
to consider efficiency, cohesiveness, and overall dialogue quality.
Automated evaluation metrics are more efficient in terms of time and resources necessary to carry
out the evaluation. Nonetheless, there still appears to be a lack of industry standards in terms of
evaluation metrics applied, and automated evaluation metrics seem to lack the ability to correctly assess
the quality, efficiency and effectiveness of the conversation as a whole. However, given the fact that
these metrics can be more easily used, they are still widely implemented to evaluate chatbots. The
evaluation metrics used to measure accuracy will be standard evaluation metrics used for Machine
Translation and other Natural Language Processing tasks such as BLEU, METEOR and TER, as they
have been used by Although these evaluation metrics are considered to be more suitable for Machine
8
Translation problems, they can still provide valuable information regarding the Textual Entailment of
the chatbot output

9
 Chapter - 3
SYSTEM ARCHITECTURE

This chapter gives the description of the system in detail. It gives details information about
proposed system with the benefits and architecture of the model.

3.1 Chat Architecture: How We Approach It

Figure 3.1 Architecture

10
 A chat consists of two major parts

• Chat App or client part, which is a Web chat application. Build on React
• Chat Server Engine or server part, which is a pool of external servers responsible for
the chatoperation. This is the place where all the chat magic happens.

Both parts contain various components that communicate to each other and bring the chat into
action

3.2 Chat App or Client Side

Chat App is the other major part of the chat architecture, the one that users directly interact
with. It's split into two separate root components:

• Chat Client Engine (fig.3.2) handles all the communication with the Chat Server Engine via
its internal components: Chat REST API Client Libraryand Chat WebSocket Client Library.
• Chat UI displays data to users: Chat Contact List UI, Chat Dialog UI

Component:
Components are the building blocks of any React app and a typical React app will have many of
these. Simply put, a component is a JavaScript class or function that optionally accepts inputs i.e.
properties(props) and returns a React element that describes how a section of the UI (User
Interface) should appear.

App.js is the starting point of our React app.A package.json file(fig.3.3):

• lists the packages your project depends on
• specifies versions of a package that your projectcan use.
• makes your build reproducible, and therefore easierto share with other developers.
A package.json file may look similar to this:

11

Figure 3.2 Package.json file of Client-side

 Figure 3.3 Client-Side Structure Tree
Figure 3.3 Client-Side Structure Tree

12
 Chapter - 4
ADVANTAGES

Faster support
Obviously chat is easy to reach for your customers, but what’s more is that the average resolution time
is significantly lower than with traditional service channels. It takes less than a minute to resolve a
customer issue using live chat.

Real-time text preview

One of the handy advantages of live chat is the option to see a real-time preview of what the customer
is typing before she hits enter. It gives your chat agent the chance to think about a solution, research
and impress users with prompt, customized answers.

Instant customer feedback

feedbackFeedback is easily collectable. Users can rate your chat service right after their interaction
with you. This is a benefit for your service agents too because they get instant feedback on their
performance. This makes it easy for them to connect the dots.

Less drama
Chat agents experience less emotional exhaustion after difficult service interactions. It’s easier to
distance yourself from a mean text than a shouting customer hurling complaints

Prevents agent fatigue

Avoid one of the main burnout reasons in customer service: boredom. By preparing templates (Chat
Macros) for frequently asked questions, chat agents have more time to focus on complex issues and
personal customer interactions.

No waiting queues
Customers reach you quicker. According to Zopim research, in less than 30 seconds. This is possible
even for small service teams because agents can handle multiple requests at the same time. Research by
Software Advice shows that the extinction of waiting queues is the primary reason why people prefer
live chat
.

13
 Chapter - 5
DISADVANTAGES

Constant monitoring

To provide the immediate responses that live chat promises, businesses must ensure that the channel is
monitored continuously when online. This can be challenging, especially for small businesses or those
operating across multiple time zones. While this is a disadvantage of live chat, a chatbot can solve this
by providing 24/7 support.

Miscommunication

The absence of verbal cues and tone can sometimes lead to misunderstandings between the customer
service representative and the customer. This is particularly true for complex issues that might be easier
to resolve through voice communication. If this is a concern for you, then adopt a live chat solution that
offers

Training

Effective live chat support requires representatives to be skilled at multitasking and typing, as well as
having in-depth knowledge of the business’s products or services. As a result, training staff to meet
these requirements can be resource intensive. To overcome this potential live chat disadvantage, ensure
you choose software that is easy to use, offers internal knowledge bases, and has AI-powered agent
support tools.

Privacy Concerns

Of all the live chat advantages and disadvantages, privacy concerns are sometimes overlooked.
Handling sensitive information over live chat requires secure, encrypted channels to protect customer
privacy. Businesses must invest in secure chat solutions to avoid data breaches and comply with
privacy regulations.

14
 Chapter - 6
APPLICATION

Chatbots are applied in many different domains. As far as Education and Research go, chatbots in
this domain seem to be mostly Information Retrieval or AIML based. Little to no Deep Learning
application have been used in these fields. The choice seems justified by the fact that chatbots created
for educational purposes are often aimed at providing specific information (such as class schedules) or
educational material. Refs. all provide examples of chatbots applied to Education and Research. For
similar reasons as in the field of education, most HealthCare oriented chatbots are Information
Retrieval based. provide different examples of chatbots applications in HealthCare.
E-commerce oriented chatbots present different configurations, mostly Information Retrieval
based configurations, but with some Deep Learning algorithms also involved in the overall
architecture. This is possibly because in e-commerce, chatbots are often used to provide customer
support. Therefore, they must be able not only to provide information on the products’ catalogue and
purchasing experience, but also conversing with the customer. provide different examples of chatbots
applied to e-commerce.
Other Information Retrieval based chatbots applications can be found in Training [Information
Technology [and Finance possibly for similar reasons.
Sequence to Sequence chatbots with attention mechanisms have been used to provide Law and
Human Resource services, respectively inIn the first case the chatbot is Long Short-Term Memory
(LSTM) based, while in the second case it is based on Bidirectional-Gated Recurrent Units. As shown,
when applied to a specific domains chatbots tend to fall back to Information Retrieval systems and
Rule-based systems, or a combination of the two. Only a few examples of applications have used
Machine Learning technologies. This can be due to several factors:
• Machine Learning in general and Deep Learning in particular, require a large amount of
training data; although training data is becoming increasingly available but finding a suitable
dataset might still represent a challenge. Furthermore, data needs to be preprocessed in order to
be used and might often contain unwanted noise.
• Training is costly in terms of infrastructure and human resources, and time consuming.
• Chatbots, when they are not used for social or companion chatbots, are usually applied
to a specific domain, which means that they require domain-specific training data (e.g.,
products information and details, financial information, educational material, healthcare
information). This type of data is often confidential due to its nature; they are not readily
available in open access to train a Deep Learning engine. Furthermore, given the nature of the
data needed and of the tasks the chatbot is required to carry out (e.g., access a customer’s
purchase history, or give more information about a product feature), Information Retrieval
might be the best solution for most use-case applications.
In conclusion, we find from the literature a clear divide in terms of chatbots’ technologies and
their application. We observe that deep Learning algorithms trained on large open domain datasets, are
usually implemented as social or companion chatbots. Task oriented chatbots appear to be usually

15
trained on smaller, domain specific, and often confidential datasets, and they are usually based on
Information Retrieval or Rule-based approaches, or a combination of both.

16
 Chapter - 7
CONCLUSION

In this paper we have provided a survey of relevant works of literature on the subject, and we
have analysed the state of the art in terms of language models, applications, datasets used, and
evaluation frameworks. We have also underlined current challenges and limitations, as well as
gaps in the literature. Despite technological advancements, AI chatbots are still unable to
simulate human speech. This is due to a faulty approach to dialogue modeling and a lack of
domain-specific data with open access. For Information Retrieval chatbots, there is also a lack
of a learnt AI model. A model like this might be used in a variety of sectors. There is still a gap
to be closed in terms of applications between industry models and current advancements in the
sector. Large models necessitate a lot of computing power and a lot of training data. There is no
universal framework for evaluating chatbots. Several models depend on human evaluation, yet
human evaluation is expensive, time-consuming, difficult to scale, biased, and lacks coherence.
A new, reliable automatic evaluation approach should be provided to overcome these
restrictions. Furthermore, recent studies have revealed a scarcity of data on the most recent
developments in language models that may be used to chatbots like Transformers. As a result,
it’s critical to examine and analyze the data used to train the various models. This type of study
provides for a more accurate comparison of different models and their results. In fact, the
distinction between chatbots’ applications and social or companion chatbots appears to be hazy.
Chatbot modeling is a fascinating challenge that mixes Deep Learning and Natural Language
Processing. Despite the fact that the first chatbots were created sixty years ago, the area has
continued to grow and provide new and exciting problems. To bridge these gaps, smaller,
flexible, less domain dependent models would be beneficial. Improved, scalable, and flexible
language models for industry specific applications, more human-like model architectures, and
improved evaluation frameworks would surely represent great.

17
 BIBLIOGRAPHY
1. Jia, J. The Study of the Application of a Keywords-based Chatbot System on the Teaching of
Foreign Languages. arXiv 2003, arXiv:cs/0310018. [Google Scholar]
2. Sojasingarayar, A. Seq2Seq AI Chatbot with Attention Mechanism. Master’s Thesis,
Department of Artificial Intelligence, IA School/University-GEMA Group, Boulogne-
Billancourt, France, 2020. [Google Scholar]
3. Bala, K.; Kumar, M.; Hulawale, S.; Pandita, S. Chat-Bot For College Management System
Using A.I. Int. Res. J. Eng. Technol. (IRJET) 2017, 4, 4. [Google Scholar]
4. Ayanouz, S.; Abdelhakim, B.A.; Benhmed, M. A Smart Chatbot Architecture based NLP and
Machine Learning for Health Care Assistance. In Proceedings of the 3rd International
Conference on Networking, Information Systems & Security, Marrakech, Morocco, 31 March–
2 April 2020; pp. 1–6. [Google Scholar] [CrossRef]
5. Kumar, R.; Ali, M.M. A Review on Chatbot Design and Implementation Techniques. Int. J.
Eng. Technol. 2020, 7, 11. [Google Scholar] [CrossRef] [Green Version]
6. Cahn, J. CHATBOT: Architecture, Design, & Development. Ph.D. Thesis, University of
Pennsylvania, School of Engineering and Applied Science, Philadelphia, PA, USA, 2017.
[Google Scholar]
7. Okuda, T.; Shoda, S. AI-based Chatbot Service for Financial Industry. FUJITSU Sci. Tech.
J. 2018, 54, 5. [Google Scholar]
8. Brandtzaeg, P.B.; Følstad, A. Why People Use Chatbots. In Internet Science; Kompatsiaris, I.,
Cave, J., Satsiou, A., Carle, G., Passani, A., Kontopoulos, E., Diplaris, S., McMillan, D., Eds.;
Lecture Notes in Computer Science; Springer International Publishing: Berlin/Heidelberg,
Germany, 2017; Volume 10673, pp. 377–392. [Google Scholar]
9. Costa, P. Conversing with personal digital assistants: On gender and artificial intelligence. J.
Sci. Technol. Arts 2018, 10, 59–72. [Google Scholar] [CrossRef] [Green Version]
10. Go, E.; Sundar, S.S. Humanizing chatbots: The effects of visual, identity and conversational
cues on humanness perceptions. Comput. Hum. Behav. 2019, 97, 304–316. [Google Scholar]
[CrossRef]
11. Luo, X.; Tong, S.; Fang, Z.; Qu, Z. Frontiers: Machines vs. Humans: The Impact of Artificial
Intelligence Chatbot Disclosure on Customer Purchases. Mark. Sci. 2019, 38, 913–1084.
[Google Scholar] [CrossRef]
12. Christensen, S.; Johnsrud, S.; Ruocco, M.; Ramampiaro, H. Context-Aware Sequence-to-
Sequence Models for Conversational Systems. arXiv 2018, arXiv:1805.08455. [Google
Scholar]