2020 8th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO)

Amity University, Noida, India. June 4-5, 2020

Survey of Various AI Chatbots Based on

Technology Used
Siddhant Singh Hardeo K. Thakur
Computer Science and Engineering, Computer Science and Engineering
Manav Rachna University, MRU Manav Rachna University, MRU
Faridabad, India Faridabad, India
[email protected] [email protected]

Abstract—Artificial Intelligence is the ability to make natural conversation. Due to some drawbacks in scripted
machines intelligent. The development in AI made machines responses, developers and researchers kept on adding new
capable to initiate the simulation of natural intelligence and are functionality and approaches in the existing ways,
able to take decisions where humans do better. Natural converging mostly to the use of Semantic Nets and Machine
Language Processing, Machine Learning are some of the major Learning to remember facts from conversation and identify
branches of AI. A Chatbot is an AI software uses Natural user intents[3]. This not only improved the efficiency of
Language Processing Techniques to simulate a chat between a Chatbots but also developed variations in work flow,
system and a user. Natural Language Processing help Chatbots technical approaches and multiple solutions to the same
to understand natural language more clearly and generate an
problem.
intelligent response. Since the introduction of first Chatbot
ELIZA in 1966, different versions with variety of technology The objective of this paper is to make a survey of
and approach have been developed. In this paper, we describe Chatbot technologies and approaches from the fundamental
the generic work flow of Chatbots and represents their model to the latest Generative model of Chatbots, so that it
comparison on the basis of technology/approach used with becomes easier for developers to understand the
some significant parameters. To make natural language infrastructure and select a technology for further
conversation more efficient, a Chatbot must analyze and development of Chatbot systems.
understand the user input correctly for a relevant response.
They are currently being used in various important domains This paper will be structured as follows: Section II
like science, education, health care etc. Chatbots have a illustrates the Architecture of Chatbot Systems. Section III
potential to improve human interaction with machines. offers the analysis of AI Bots which is followed by sub-
section A discusses about History of Chatbots and sub-
Keywords—CHATBOTS; AI; ML; NLP; Deep Learning; QA section B which provides survey of various Chatbots based
Systems. on their implementation techniques. Section IV discusses
about the Techniques and Approach opted to develop
I. INTRODUCTION Chatbots. Section V represents the Comparison of Chatbots
Today there is an epoch of Artificial Intelligence. The and Section VI will conclude the overall paper.
science behind human thinking and acting developed by
psychologists and philosophers, welcomed by technological II. CHATBOT-SYSTEM ARCHITECTURE
experts with robust sophistication developed the science of It is important to understand the architecture of Chatbot
Artificial Intelligence[1]. Natural Language conversational before development. The correct architecture depends upon
entities powered by AI, called Chatbots are the main focus of
specific domain of Chatbot but the basic communication
this study. The term ‘Chatbot’ was coined by Michael
flow remains same. Fig. 1 explains the fundamental work
Mauldin in 1994 comprises of two terms ‘Chat’ and
‘robot’[2]. Chatbot is a software entity developed to mimic flow of Chatbot. Chatbot takes input from user. The mode of
the natural conversation with human partner. Chatbots are communication can be text, voice based or both. The set of
powered by AI uses Natural Language Processing to instructions or a query is then passed to Natural Language
understand natural language. They are very much popular Processing Component in a form of text. It uses context
these days because of various success stories and information to classify intents and entities from the input.
controversies created, including Amazon’s ALEXA, Apple’s By remembering facts from previous conversation it
SIRI and Microsoft’s TAY. Today most of the Chatbots are becomes easier to find out the accurate intent of user input.
accessed via messaging apps such as FB Messenger or Response Generator uses different algorithms, processes
WeChat, via virtual assistants such as GOOGLE predefined database and calls an external APIs also if
ASSISTANT, etc. It has a future to automate customer required to process user request. It perform all possible set
services and reduce human efforts. of calculations to generate domain specific response. The
The biggest challenge that the majority of existing result of these calculations is a set of responses which is
Chatbots have is to understand human inputs and its further passed to Response Selector. Response Selector uses
responses with emotions. By making Chatbots empathize ML plus AI algorithms to select response. The selected
with people’s emotion, can make the users feel more response is further optimized by Message Generator which
understood and satisfied. Pattern Matching and Parsing uses some placeholders to generate output which would
algorithms are the common techniques used and these work better for the user.
approaches however not able to simulate a satisfactory

978-1-7281-7016-9/20/$31.00 ©2020 IEEE 1074

Authorized licensed use limited to: University of New South Wales. Downloaded on November 02,2020 at 19:43:50 UTC from IEEE Xplore. Restrictions apply.
Fig. 1. System Architecture Of Chatbots

parsing but has no built-in module/functionality for

III. ANALYSIS OF AI BOTS contextualizing-events. By using these methodologies it is
able to create an illusion of understanding and convinced
A. History Of Chatbots most of the users for being able to pass Turing Test.
In 1950, Alan Turing published his paper “Computing However, the development of ELIZA is not considered to be
Machinery and Intelligence” which considers the question a failure. It is a great source of motivation and knowledge
“Can Machines think?” and introduced Turing Test for for innovators focused on AI and NLP.
estimation and evaluation of intelligence simulated by
After ELIZA which simulated Rogerian therapist,
machines. Turing Test is the method of estimation in AI to
PARRY attempted to figure out a person with paranoid
check whether a computer is capable enough to think like
schizophrenia. It was developed by Kenneth Colby in 1972.
human being or not[4].
It uses the same approach used in ELIZA but it is much
ELIZA is considered to be the first entity in a world of more advance than ELIZA. It used complex system of
Chatbots developed by Joseph Weizenbaum in 1966 at MIT assumptions ,attributions and emotional responses to mimic
AI Laboratory. Weizenbaum developed ELIZA to simulate the behaviour of paranoid individuals for health care
the behaviour of Rogerian Psychotherapist(a psychotherapy analysis. In 1970, PARRY undergone evaluation by a group
developed by psychologist Carl Rogers). ELIZA is the first of psychiatrists and able to qualify with no identification by
program capable of attempting the Turing Test[5]. After 52 % of time[6].
ELIZA Kenneth Colby in 1972 created PARRY, a program
JABBERWACKY is a Chatbot created by Rollo
identified as ELIZA with attitude. It simulate a person with
Carpenter in 1982. It is considered to use an AI technique
paranoid schizophrenia and used same approach with more
called contextual pattern matching. It simulate natural
advancement to that used by Weizenbaum[6]. It was tested
human chat in interesting, entertaining and humorous
using Turing Test with some variations in 1970s and
manner and the first Chatbot to accept voice inputs[7].
correctly identified nearly 48 percent of the time. However,
Whereas all the previous Chatbots had relied on static
the first Chatbots were based on identifying keywords from
database, JABBERWACKY collects tokens or group of
the user inputs and applying rules to transform keywords in
keywords from the set of previous and current conversations
resulting comments. If the keywords are not found, they
and dynamically grow its own content by adding them in
reply by returning a generic output from previous
database. Rollo developed JABBERWACKY with the
conversation. These Chatbots used pattern matching and
purpose to create AI that is able to pass Turing Test.
parsing methodology that gave interrogators an illusion of
However, it is a two time Loebner Prize Winner(2005 and
natural conversation and intelligence. With minimal
2006). The new version of JABBERWACKY came in 2008
contextual understanding, these rudimentary models passed
and now it is popular by name CLEVERBOT.
the Turing Test which raised arguments on it. This causes
slight changes as an improvement in Turing Test. Inspired by Weizenbaum’s ELIZA, Richard Wallace
developed A.L.I.C.E stands for Artificial Linguistic Internet
B. Survey Of Chatbots And Their Implementation Computer Entity or simply ALICEBOT in 1995. ALICE is a
Techniques natural language processing Chatbot based on heuristic
ELIZA is an early NLP Chatbot created by Joseph pattern matching technique to evaluate response[8]. The
Weizenbaum at MIT AI Laboratory. Weizenbaum program uses an XML dialect called Artificial Intelligence
developed ELIZA to study the conversation between man Markup Language popular as AIML which allows a wider
and machine[5]. It uses pattern matching and some sort of bank of more sophisticated responses. It is one of the highly

1075

Authorized licensed use limited to: University of New South Wales. Downloaded on November 02,2020 at 19:43:50 UTC from IEEE Xplore. Restrictions apply.
efficient programs of its type and secured Loebner Prize by Microsoft in 2007 and are integrated with a Semantic
three times(2000, 2001, 2004). But it is based on predefined search Database called Satori[13]. It consists of a speech
set of responses to generate output. However, it is not able to enabled channel operates on voice inputs and outputs. It can
pass Turing Test due to some drawbacks. set reminders, recognize natural voices and can answer
questions using information and web results from the Bing
SMARTERCHILD is an intelligent Chatbot developed search engine.
by ActiveBuddy. It was released in June 2001 by using AIM
platform which is now discontinued. SMARTERCHILD is a After few months of release of CORTANA, Amazon
Chatbot designed to provide quick access to news, weather, introduced its new iteration of virtual assistant that is
stock information, movie listings, and transit schedules. It ALEXA. ALEXA uses NLP algorithms for accurate voice
has an add on functionality to be used as personal assistant, interactions. It is capable of setting alarms, playing music
calculator and translator[9]. This type of bot completely playlist, setting up reminders, providing news, playing audio
changed the scenario of interaction with machines. It also books and streaming pod casts[15]. It can interact with
initiated the era of having virtual assistants. Smart Devices and used as hub to implement Home
Automation. AMAZON had done business of 100 million
Later on, in 2006 IBM introduced WATSON. WATSON ALEXA - enabled smart devices in the entire world,
is a QA computer system developed using IBM’s DeepQA reported by AMAZON’S team in Jan. 2019 .
software and Apace UIMA framework implementation. It is
implemented on the SUSE Linux Enterprise Server 11 OS The another noteworthy AI Chatbot that came into
and Apache Hadoop framework to provide distributed existence is TAY. Microsoft Corporation released TAY via
computing. WATSON has an ability to run hundreds of Twitter on March 2016. It was designed to mimic the 19
language analysis algorithms on same time and shows high year old American girl and made public as Twitter bot. To
level of linguistic intelligence[10]. It can access enormous make TAY learn from interaction with public is a major
databases of knowledge and can access 200 million pages of approach to expand knowledge base of developers focused
data in short while making it the ideal QA system. on AI. But Microsoft also shut down the service only after
WATSON participated in Jeopardy competition and won in 16 hours of its launch because of posting offensive tweets
2011. WATSON ’S capabilities have been extended in related to racism ,feminism etc[16]. Just like human kids it
recent years. Therefore, IBM WATSON is also considered learns and mimics what it is hearing and easily get exploited.
as the brain or information store for many conversational The problem is that Microsoft dropped this little kid among
entities. It is used in health care, as teaching assistant, in a bunch of crazy thugs and trolls with a very little protection
weather forecasting, fashion, advertisements,etc. and had not given it the understanding of moral values.
However, TAY was replaced by ZO, an English language
SIRI is a virtual assistant released in 2011 by Apple Inc. Chatbot developed by Microsoft. It was the English version
It is available on many platforms including Apple Inc’s iOS, of Microsoft’s other successful Chatbots XIAOICE (China)
watchOS, iPadOS and tvOS operating systems. The assistant and RINNA (Japan). But due to some controversial reports,
uses ASR(Automatic Speech Recognition) and advance it was shut down on multiple platforms.
machine learning algorithms to translate human speech into
text, answer questions, make recommendations and perform Looking forward, Google developed AI enabled
actions[11]. It validates a wide horizon of user inputs GOOGLE ASSISTANT in 2016. GOOGLE ASSISTANT is
including performing system actions, checking knowledge an extension of GOOGLE NOW, programmed in C++ and
base, scheduling events and reminders for curriculum, unveiled in May 2016[17]. It supports both voice and text
controlling device settings, searching Internet etc. It has a based interaction. It is able to access APIs using internet, set
functionality to forward the user input unable to answer to a reminders, schedule events and alarms, etc. GOOGLE
more generalized QA services. SIRI is released to support ASSISTANT facilitates the action of smart home control. It
many natural languages, praised for its voice recognition and is operative with a large store of devices, so that you can
contextual knowledge. control home appliances with your voice. But it does not
work without Internet .It also suffers from maximum battery
MITSUKU is an AI Chatbot created from usage and heat up your mobile phone.
AIML(Artificial Intelligence Markup Language) by Steve
Worswick in 2012. It is considered to be an advance Instead of humans learning how the machine interacts, it
intelligent system rectified the major drawbacks seen in is the machine that needs to learn and adapt to us. Working
most of the Chatbots. To simulate a natural chat it on this conceptual philosophy, Samsung released voice
remembers the facts from conversation and avoid answering powered digital assistant named BIXBY in 2017[18]. Using
the same questions again and again[12]. It is built using Deep learning concepts in a core of user interface, Samsung
A.L.I.C.E. AIML files with many integration based on introduced intelligent interface in BIXBY. Completeness,
conversations. It has been taken 9 or 10 years to develop a Context Awareness, Cognitive Tolerance are the three major
record breaking five time Loebner Prize winner(in 2013, properties of BIXBY. It is expected that it will change the
2016, 2017, 2018, 2019). experience of interaction with Smart-phones.
CORTANA is a virtual assistant developed by Microsoft The comparison between Chatbots of different timeline
in 2014 and made available in 2015 for Windows. Currently surveyed above is represented in Table 1.
it is available in English, German, French and 6 more
language editions for Android, Windows, iOS and Xbox OS IV. TECHNIQUES AND APPROACH TO DEVELOP
like popular Operating Systems. Windows 10, Windows 10 CHATBOTS
Mobile, Microsoft Band, Amazon ALEXA etc are the
platforms to access CORTANA[14]. The NLP functionality 1) Rule Based Conversation: It is a very basic technique
of CORTANA are accessed from Tellme Networks owned used in most of the Chatbots. It consists of predefined

1076

Authorized licensed use limited to: University of New South Wales. Downloaded on November 02,2020 at 19:43:50 UTC from IEEE Xplore. Restrictions apply.
hierarchy of rules that are used to transform user input into drag and drop approach. There are various AI Bot builders
an output. The set of rules can be simple or complex breaks available including Mobile Monkey, Botsify, Chatfuel,
down the input into sequence of tokens to identify SnatchBot etc to develop Chatbots in limited time and cost.
pattern[19]. The decomposition of text into set of words is
significant to understand the grammatical structure of an V. COMPARISON OF CHATBOTS
input[20]. The classified set of keywords or patterns are Table 1 shows the timeline of Chatbots from beginning.
used to generate output. The first Chatbot ELIZA is based It represents the comparison of Chatbots on the basis of
on this approach. However, these Chatbots are not able to some significant parameters and gives them a sort of
answer patterns that don’t match predefined script. feedback on technologies used. From the comparison table it
is found that the development and improvement of Chatbots
2) Deep Learning: In this method, Chatbots are created
is not grown at an expected rate. It is observed that there are
using Machine Learning Algorithms. A Deep Learning multiple solutions and approach to the same problem. The
Chatbot learns everything from its data and conversation latest Chatbots are nowadays based on Generative approach.
with human[21]. It is trained to develop its own opinion on Generative Models are more prompt to research these days.
text and can be more effective by using more and more However, developers are showing keen interest in Python as
data. RETRIEVAL BASED and GENERATIVE are two a platform to build Chatbots because of the presence of
major types of bots that can be created using Deep AIML, simplicity and versatility. Python is an open source
Learning. Retrieval Based Bots uses heuristics and semantic programming language with great community, reduces code
net to estimate the most accurate response from the and provide lots of predefined libraries to implement AI
predefined repositories but Generative model is an advance conversational entities.
form doesn’t use any type of repository. Generative
Chatbots can build a response they have never seen before. VI. CONCLUSION
Generative models generate mistakes and difficult to be In this paper, we have covered a number of significant
trained. On other hand, Retrieval based Chatbots also papers and reference material related to Chatbots and their
returns incorrect outputs because they are totally based on implementation techniques. They are programmed to
retrieval of data[22]. simulate natural conversation like human being but remains
3) Ensemble Methods[23]: The modern Chatbots like short of being able to pass Turing Test. A comparison has
been made to identify the most efficient technical approach
Alexa nowadays, designed to communicate like a virtual
for simulation of natural conversation. From the survey
family member. They have been built using Ensemble above, it is found that the Generative Models based on Deep
Methods which inherits the functionality of Rule Based, Learning can be very effective for building Chatbots. These
Retrieval Based and Generative methods to process user type of Chatbots are also not being able to pass Turing Test
request. The Chatbots built using this approach are capable but a proper training using huge amount of quality data and
enough to simulate conversation on unspecified subjects. combination with Retrieval Based approach can yield better
4) Domain Specific Chatbot: The approach to develop results. IBM WATSON is a computer system with enormous
Domain Specific Chatbot improves the efficiency of database can be used as a teaching assistant for Generative
Chatbots. Such Chatbots are used in variety of domains Chatbots.
including Education, Health care etc. This approach allows Currently Chatbots have limited linguistic intelligence.
Chatbot to implement the above mentioned techniques and So, there is a great scope for removing such language
methods for a particular domain. It helps Chatbots to extend barriers in future Chatbots by making it capable to
the coverage area in a particular subject. understand multiple languages or dialects.To simulate a
5) Chatbot Builders: Chatbot Builders are the tools to natural conversation with user the current Chatbots need to
create Chatbots without coding. There are ready made learn emotions to identify actually the entire possibilities of
solutions and predefined scenarios that can be used to user input.
develop Chatbots in limited time period. It simply works on

TABLE I. COMPARISON OF CHATBOTS

YEAR OF CHATBOTS DEVELOPE PROGRAMMIN TECHNICAL CONTEXTUAL- MODE OF TYPE
RELEASE D BY G/PLATFORM DESCRIPTION/WORKFLOW UNDERSTANDI INTERACTI
USED NG ON
1966 ELIZA- [5] Joseph MAD-Slip User input is analyzed according to No Text RULE BASED
Weizenbaum the set of decomposition rules to
identify keywords. Responses are
generated by reassembly rules and
pattern matching techniques.
1972 PARRY - [6] Kenneth Colby Assembly User input is analyzed to identify No Text RULE BASED
Language, M-Lisp tokens. Complex system of
assumptions, attributions and
emotions is used to generate
responses to mimic the behaviour of
paranoid individuals.
1982 JABBERWACK Rollo Computer Natural Language Processing and Yes Text RULE BASED
Y - [7] Carpenter Programming (not Contextual Pattern Matching is
known) used to analyze user input and
generate responses. It learns by
associations, collect phrases used in

1077

Authorized licensed use limited to: University of New South Wales. Downloaded on November 02,2020 at 19:43:50 UTC from IEEE Xplore. Restrictions apply.
 inputs and store it in dynamic
database.
1995 ALICE - [8] Richard JAVA, AIML Natural Language Processing No Text RETRIEVAL
Wallace Chatbot simulate conversation with BASED
human being using AIML based
heuristic pattern matching
techniques.
2001 SMARTERCHIL ActiveBuddy C++, Adobe Based on Natural Language Chat Yes Text RETRIEVAL
D - [9] Inc. Flash(by Functionality includes wide range BASED
assumption) of database applications(eg.Sports,
Weather, Movies)
2006 WATSON - [10] IBM JAVA, C++, High Level Linguistic Intelligence Yes Text RETRIEVAL
Prolog, Apache system runs hundreds of algorithms BASED
UIMA simultaneously to analyse natural
-SUSE Linux language. Served with complex
Enterprise Server database can quickly access 200
11 million pages of data.
2011 SIRI - [11] Apple Inc. Objective C On receiving request, SIRI Yes Text + Voice RETRIEVAL
translates and breaks down the BASED +
code to analyze particular GENERATIV
patterns, keywords and E
phrases. The set of data then
runs into an algorithm and
tested with number of
combinations of sentences to
identify the exact meaning of
input. Then it begins to assess
what task needs to be carried
out and complete the sentence
into a response.
2012 MITSUKU - Steve AIML It is developed with AIML Yes Text RETRIEVAL
[12] Worswick technology. It takes user input, BASED +
identifies the subject in input GENERATIV
and queries related to it and E
returns response. It has an
ability to maintain reasons with
specific objects.
2014 CORTANA - Microsoft C#, C, C++ CORTANA uses information Yes Voice RETRIEVAL
[14] from Bing Search BASED
Engine(weather, sports, traffic,
etc) to replicate natural
conversation. It takes input,
analyze and perform actions
based on pattern matching and
information from search
engine.
2014 ALEXA - [15] Amazon Node.js, JAVA, ALEXA takes input, breaks Yes Voice RULE
PYTHON down the order to analyze the BASED+
task and carry out the RETRIEVAL
corresponding functions. Calls BASED +
API if required and replies GENERATIV
back by performing actions. E
2016 TAY - [16] Microsoft Python, Java, It was designed to learn from Yes Text RULE
Research, Node.js its surroundings. Based on the BASED +
Bing interactions made with people, GENERATIV
it made its own opinion. But E
the trolls on social media
caused very offensive impact
on it.
2016 GOOGLE Google C++ It uses Smart speaker Yes Voice + RETRIEVAL
ASSISTANT - technology to accept input, Text BASED +
[17] NLP to analyze text, invokes GENERATIV
Actions SDK to fulfill request. E
Actions SDK generates
response corresponding to your
reuest.
2017 BIXBY - [18] Samsung Samsung BIXBY is divided in 3- Yes Voice + Text RETRIEVAL
Development Kit subparts BIXBY voice, BASED +
BIXBY Vision, BIXBY Home. GENERATIV
BIXBY voice consists of the E
use of smart speaker to call
service by voice. BIXBY
Vision is an augmented reality
camera used for to carry out
image processing tasks.
BIXBY Home consists of the
list of menus on which BIXBY
can interact with.

1078

Authorized licensed use limited to: University of New South Wales. Downloaded on November 02,2020 at 19:43:50 UTC from IEEE Xplore. Restrictions apply.
 [9] “BotSociety Blog- SmarterChild And Siri- A Twisted Love Affair”.
ACKNOWLEDGMENT [Online]. Available :
https://botsociety.io/blog/2018/02/smarterchild-siri-a-twisted-love-
This research is supported by Manav Rachna University. affair/
We thank our colleagues from Manav Rachna University [10] “The History And Future Of Chatbots-Posted By Kate Austin Sept.
who provided the insight and expertise that greatly assisted 21, 2017:- InTheChat”. [Online]. Available :
the research, although they may not agree with all of the https://inthechat.com/the-history-and-future-of-chatbots/
[11] SIRI Virtual Assistant. [Online]. Available:
conclusions of this paper. “https://en.wikipedia.org/wiki/Siri”
We thank Dr. Hardeo K. Thakur, Associate Professor Of [12] Steve Worswick Interview - Loebner 2013 winner. [Online].
Available:
Computer Science for assistance and comments that greatly “https://aidreams.co.uk/forum/index.php?page=Steve_Worswick_In
improved the manuscript. terview_-_Loebner_2013_winner#.XhQMJfzhVPY”
[13] Cortana(Intelligent Assistant):- ISJERT Vol.6 Issue 4, April 2017,
We also thank 3 “anonymous” reviewers for their so- ISSN:2278-7798, International Journal Of Science, Engineering
called insights and grateful comments. And Technology Research.
[14] Cortana’s Regions And languages- Microsoft Retreived Sept.18,
2016.
REFERENCES [15] “Alexa Voice Service Overview (v20160207)|Alexa Voice Service”.
[1] International Symposium on Electronics And Smart Devices-Brain [Online]. Available : www.developer.amazon.com.
Inspired Cognitive Artificial Intelligence For Knowledge Extraction [16] Wakefield, Jane.”Microsoft Chatbot is taught to Swear On Twitter”
And Intelligent Instrumentation System, 2017 BBC News 25 March, 2016.
[2] Chatbots-An Interactive Technology For Personalized [17] Google Assisstant. [Online]. Available:
Communication, Transactions And Services - IADIS International “https://en.wikipedia.org/wiki/Google_Assistant”
Journal on WWW/Internet Vol-15, No.1 ISSN:1645-7641, 2017 [18] Bixby : A New Way to Interact with Your Phone (Samsung News
[3] A Survey Of Chatbot System through Loebner Prize Competition- Room). [Online]. Available:
Artificial Intelligence Laboratory, 2012 https://news.samsung.com/global/bixby-a-new-way-to-interact-
[4] “Turing Test”, Turing, Alan- Computing Machinery And with-your-phone
Intelligence, 1950 [19] Pattern Matching. [Online]. Available:
[5] “ELIZA”, Weizenbaum, Joseph(1976), Computer Power And https://en.wikipedia.org/wiki/Pattern_matching
Human Reason: From Judgement To Calculation, New York: W.H. [20] Parsing. [Online]. Available : https://en.wikipedia.org/wiki/Parsing
Freeman And Company.pp.2, 3, 6, 182, 189, ISBN: 0-7167-0464-1, [21] Deep Learning Chatbots. [Online]. Available:
1976. https://hackernoon.com/deep-learning-chatbot-everything-you-need-
[6] “PARRY: The AI Chatbot From 1972 - Phrasee”. [Online]. to-know-r11jm30bc
Available : https://phrasee.co/parry-the-a-i-chatterbot-from-1972/ [22] Deep Learning for Chatbots. [Online]. Available:
[7] “Jabberwacky” IEEE 802.3 12.4.3.2.3 Jabber function, 2018. http://www.wildml.com/2016/07/deep-learning-for-chatbots-2-
[8] A Comparative Study Of Chatbot Implementation Techniques, retrieval-based-model-tensorflow/
ISSN 2278-0882 Volume 7, Issue 10, International Journal Of [23] 6 Technical Approaches For Building Conversational AI. [Online].
Scientific Research Engineering And Technology (IJSRET), 2018. Available: https://www.topbots.com/building-conversational-ai/

1079

Authorized licensed use limited to: University of New South Wales. Downloaded on November 02,2020 at 19:43:50 UTC from IEEE Xplore. Restrictions apply.