Proceedings of the Fifth International Conference on Trends in Electronics and Informatics (ICOEI).

IEEE Xplore Part Number:CFP21J32-ART; ISBN:978-1-6654-1571-2

Healthcare Chatbot System using Artificial Intelligence

* * *
Nikita Vijay Shinde , Aniket Akhade , Pranali Bagad
sniki241199@[Link] aniketakhade74@[Link] pranalibagad1410@[Link]

* * *
Harshit Bhavsar , Dr. [Link] , Prof. Amol Kamble ,
bhawsarharshit9691@[Link] sharmila.wagh123@[Link] [Link]@[Link]

Abstract concept. After knowing about the word chatbot

This paper presents Healthcare Chatbot many questions come to mind like what is it? How
using Artificial Intelligence that can make a does it work as a human? Where can we use it? and
human-system interaction to resolve basic why is it useful for the healthcare sector? Basically,
queries regarding health parameters before Chatbot is a program which is created for
consulting a doctor. The actual purpose communication between human and machine via
behind this work is to work on the user’s text. It’s software which helps people to make
2021 5th International Conference on Trends in Electronics and Informatics (ICOEI) | 978-1-6654-1571-2/21/$31.00 ©2021 IEEE | DOI: 10.1109/ICOEI51242.2021.9452902

symptoms and to provide medical suggestions coherent conversation using NLP in different
according to it, to reduce the time and cost languages. For example, we can say Amazon AI’s
required for the process. It is a system that like Alexa, Siri but they are digital voice assistant
communicates with users using Natural and a chatbot has text-based interaction with users.
Language processing rather we can say it is a These systems gain knowledge using web resources
computer program that uses NLP for or with human assistance. In the current pandemic
interaction. The chatbot works on provided situation going to a hospital is a high-risk activity so
input by the user, It takes sentence keywords our chatbot will provide necessary guidelines and
and makes decisions to solve the user's query medical recommendations to the user according to
and answers it accordingly. Some of these their health conditions and current symptoms. In our
calculations like Rank calculation and paper, we are trying to develop a healthcare chatbot
sentence similarity are performed with the that will help people who need medical assistance at
help of TF-IDF, Stemming, n-grams and home.
cosine similarity. Basically, our system uses
machine learning techniques to solve user's II. RESEARCH GAP
queries regarding healthcare by using
suitable algorithms. The system allows users The chatbot used by the various sectors for services
to create their profile to specify their is kind of scripted. They help their users with some
symptoms, to suggest doctors and dosage previously defined queries. Research is being
reminders. This chatbot can be used by carried out in the development of standard tedious
normal humans in any kind of emergency chatbots to be active and responsive and carry out
situation where it can play the role of adviser the communication in a normal/natural language.
to people about primary care before This all process requires the involvement of
consulting a doctor or sometimes it will work Machine Learning and Natural Language Processing
as a doctor for minor and short term health techniques in the system. There are many ways to do
issues like cold, headache, etc. Along with this so. Selecting a significant technique is completely
chatbot will be a support for the needy who based on the domain of the chatbot, the features it
require urgent solutions. A user can be able to intends to provide, the language used for
recognize the actual disease by providing communication, the user, etc. All the issues
symptoms of it. As if a person will also know mentioned above need to be examined while
about the solutions or we can say precautions working on the implementation of chatbot.
and remedies that they should take
accordingly. III. LITERATURE SURVEY

In this paper the study is based on a chatbot

Keywords :- Artificial Intelligence, Healthcare, healthcare system using AI. This paper helps the
Chatbot, Medical Assistance, NLP, TFIDF. user to solve their queries regarding small/minor
health [Link] UI and Architecture of the
I. INTRODUCTION proposed system are very short and simple. They
provided a register and login page to the users. If the
As we know, nowadays computers are helping chatbot couldn't give the answer then with the help
humans in every single way. In computer of the database expert system gives answers to the
technology, Artificial intelligence has been query. In the proposed system, users have to ask a
introduced to improve systems in many different query then that query is served by a [Link]-IDF,
ways. A lot of people are working on AI to create Stemming, N-gram, and cosine similarity are used
artificial human brains that can think and respond properly. [1]
like a human. A small example of this technology is
a chatbot. As it sounds it is really an interesting

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
978-1-6654-1571-2/21/$31.00 ©2021 IEEE 1174
The paper describes the detailed use of various experimental results using score graphs. The author
algorithms like N-gram, Stemming, TF-IDF, and conveys that the approach used by him is suitable for
cosine similarity. This paper states how one can use any domain for chatbot which will upgrade the
these algorithms so that he could get an optimized productivity and performance of the chatbot. [7]
result quickly. It describes various queries and ways
to handle them by chatbot. For convenience they This research came up with development of a
also gave test cases for algorithms and also chatbot system with an emotional model/system. To
explained functional Architecture of databases. understand human interactions, computers need to
Basically, chatbot quality and [2] make extra efforts. In this paper, emotions,
sentiments and personality factors are implemented
Here In this paper, chatbot is created for the in the chatbot. They have used several public
healthcare purposes, which provides a healthcare datasets and AIML techniques at a time of
assistant to the user. A user can interact with a implementation of a model and performed five
healthcare assistant using a dialog interface. It gives phases in the research from which we collected some
functionalities like disease detection from user's ideas to improve the chatbot performance in the
symptoms, medical term description, doctors healthcare domain also. Basically the proposed
suggestions, Scheduling treatments, system was a kind of simple interview. They have
tracking/monitoring user's health parameters. They shown their experimental results using statistical
have checked overall performance in both analysis format. [8]
environments i.e. Offline and Online. Also they
performed many experiments to check the system IV. SYSTEM ARCHITECTURE
quality/performance. In the last few years, many
literature work has been done on this topic to create The Figure 1 shows the architecture of the chatbot
an eHealth environment for the ease of patients or system.
we can say for patients convenience. [3]

In this paper, they search on all chatbots and

technologies used in chatbots to prove that how we
can trust chatbot. Chatbots are useful in the current
world and business. Users can feel comfortable
chatting with chatbot too. ELIZA was the first
experiment of chatbot with successful experience.
There are many more chatbots explained over here
like PARRY, ALEXA etc. For that they used
Artificial intelligence, machine learning, BRNN and
condor tribe finder systems. They also explain how
these technologies play an important role in the
chatbot system. Chatbot performance measurements
are given over here which is nothing but
Conversation length and structure, retention rate,
ability to provide personalized communication etc.
These are the measurements which are currently to
measure performance of chatbot. In any chatbot,
while creating it trust matters the most. Chatbot must
have to communicate in such a way that the user has
to feel friendly and comfortable with it. Especially
in healthcare, trust on chatbots has to be created
because the user or patient shares personal Fig 1: System Architecture
information and it required to be safe. [4]
1] Tokenization
In this paper, they created a chatbot system with Tokenization is the process of separating
hardware and software. This chatbot runs with sentences, raw text or documents in the form of
Bluetooth and also does movements. They used tokens. Tokens are nothing but words or characters
voice communication for commanding chatbots. or subwords. This is done mainly to increase the
The query asked by the user will be searched in a processing. These tokens are considered a building
database that is stored in raspberry pi. This chatbot block of natural processing language.
works only with specific diseases like cold, typhoid, 2] Stop word removal
malaria etc. [5] When we process raw data, it may contain
lots of words that are not necessary for the
This paper shows the performance and model processing purpose so while performing a search
building for a healthcare assistant(chatbot) in deep operation for a query, those words should be filtered
learning. In this paper, they are using deep learning out. Words like "the", "an", "in" and so on. are stop
technology with Bidirectional RNN. This words. They keep unnecessary impact in statements,
technology is supported by an attention mechanism. so that words should be removed.
They have developed a model to execute English to 3] Stemming and Lemmatization
English translation. The paper specifies a major role Stemming is nothing but reducing a word into its
of attention mechanism. The experiments were stem word that is the root word. Root words are
organized using Tensorflow and showed nothing but the word from which the other word is
Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
derived. For e.g. Game is root word and Gaming, result of similarity calculation ranges from 0 to 1
gamer are derived words. Stemming and and Frequency can never be negative. The cosine
Lemmatization are almost the same. The small similarity can be calculated as follows:
difference between them is that lemmatization gives 𝐴.𝐵 ∑𝑛𝑖=1 𝐴𝑖 𝐵𝑖
𝑠𝑖𝑚𝑖𝑙𝑎𝑟𝑖𝑡𝑦 = 𝑐𝑜𝑠(𝜃)= =
more accuracy than Stemming. In this way, ||𝐴|| ||𝐵||
√∑𝑛𝑖=1 𝐴2𝑖 √∑𝑛𝑖=1 𝐵2𝑖
stemming reduces the size of the index and increases
retrieval accuracy.
4] Tf-IDF – TFIDF stands for Term Frequency and Equation 1
Inverse Document Frequency. With the help of this,
the weight of each and every word in the document Where, ||A|| is Euclidean norm of vector A and ||B||
and corpus is calculated. This helps to determine Euclidean norm of B. The measure calculates the
how much is important. cosine angle between vectors A and B. The cosine
𝑐𝑜𝑢𝑛𝑡 𝑜𝑓 𝑡 𝑖𝑛 𝑑 similarity is a number between 0 and 1 and is mostly
tf(t,d) =
𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑤𝑜𝑟𝑑𝑠 𝑖𝑛 𝑑 used for plagiarism detection.
IDF - IDF stands for Inverse document frequency
and its measures informativeness of term. document
frequency is nothing but the occurrence of the term
in the document and which is given by,
𝑁
idf(t) =
𝑑𝑓
IDF is very low for stop [Link] some cases, there
will be a large corpus so we take log of IDF value.
𝑁
Idf = 𝑙𝑜𝑔 ( )
𝑑𝑓
Therefore, weight of each term in the document can
be calculated as,
𝑁
Wi = tfi*𝑙𝑜𝑔 ( )
𝑑𝑓
5] N-gram
An n-gram is a sequence of n items which are
neighbors (words, letters or symbols) in a document.
Using these n-grams and the possibilities of the
occurrences of exact words in exact sequences are
calculated so ultimately it improves the projection of Fig 2 : System Flow
autocompletion systems.
As we can see in this flow diagram, processing of
6] Sentence similarity: Cosine similarity is a system is presented. Use of directed arrays shows
technique which is used to check whether the two the flow of process which is explained below with
sentences are similar or not. The similarity between examples.
the present query and the document is directly Fig 2 shows the system flow.
proportional to the number of weight of query. The

Remove all punctuation marks from a sentence or we can say all the stop words are removed from the sentences
which makes the process easy to process for further steps.
Here in the below figure we can see that, the underlined punctuations are removed after the process.

Fig 3 : Removing Punctuations

Tokenizing means splitting a statement, paragraph or a whole document into words . Those words or smaller
units known as tokens. Here, In the below image we can see a tokenized text document.
Here, In Fig 4 we can see a tokenized text document

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
 Fig 4 : Tokenization

Stemming is a technique which is used to extract the actual / base form of the words by removing affixes. In the
below image we can see that the words lives and joking are extracted into their base forms live and joke
respectively.
In Fig 5 we can see that the words lives and joking are extracted into their base forms live and joke respectively

Fig 5 :Stemming

TF—IDF means Term Frequency Inverse Document Frequency. This is a concept which uses to measure the
quantity of word in documents. Basically, it computes a weight of a single word which implies the importance of
that word in the document. In short, this method is used in the term text mining. In fig 7 output of an example is
shown.

Fig 7 : TF-IDF

Computing the similarity between texts is necessary. Search engines required to simulate the significance of a
document to a query. In fig 8, it shows the similarity result.

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
 Fig 8 : Similarity

V. RELATED MATHEMATICS f4(Final_textToInterface) :Si → Ri, It

results given Text to Result_text(Rtext).
Let S is our proposed system
S = {I, O, F, Fa / ϕ }
Where, ϕ is constraints of the system
● I is an input I = {T, Ts} ϕ = {VI, TD, IC}
where, where,
T = Input Text VI = Valid Input (Text)
Ts = {Set of Training Set Data} TD = Correct Training Data Set
● O is an output, O = {Qr} IC = Internet Connection.
where, Result :
Qr = Response to query Fig 6 shows the expected result. After processing
● F is failure of the system, all above operations, the final answer of the query
F = {WLI, WQr, WI} of the user is obtained. In simple words, It catches
where, keywords from the user’s queries, starts performing
WLI = Wrong Input in case of with the system flow and generates the final result
Language on the chatbot interface.
WQr = Wrong response to query
WI = Wrong Input by user (like
special character)

Function set : Fa = {f1, f2, f3, f4}

where,
f1 = TextToText
TextToText (T)
Input = Text (T)
Output = Response(R)
Failure = Wrong or null
Output
f1 (TextToTokenization) : Ti → Ni, It
converts Text to Tokenization (Ni).
f2
(Tokenized_textToRemoving_punctuation
) : Ni → Pi, It converts Tokenized Text to
Remove Punctuations (Pi).
f3 (Removing_punctuationToStemming) : Fig 6 : Chatbot Interface
Pi → Si, It forwards Text for Stemming(Si).

Detection Test Using General Word Percentage Corpus GWP Terms Terms Term
Number Ratio Present Detected Detection
Approach : Ratio
1 0.923 1 2 0.500
Here, we take some sample to test. We would
check that our Chatbot detects medical 2 0.800 3 1 0.330
nomenclature messages for our sample:
1. Hello bot! I am feeling well today! And I am 3 0.900 1 0 0.000
having a severe headache.
2. I can't deal with this headache, also I feel like 4 0.860 3 3 1.000
vomiting. I think My digestion is not okay.
3. Is brain cancer curable? I think I am suffering 5 0.200 1 1 1.000
from it.
4. I think I have typhoid. Average 0.736 1.8 1.4 0.566
5. I am feeling like vomiting.

Table 1 : Illustration of term detection ratio GWP Ratio= Total Number of unrelated medical
Ratio Calculation words / total number of words

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
SR. Questions Possible Answers
NO.
1 What is your Male Female
gender?
2 What is your age 18-25 26-35 36-45 46-55 Over 55
range?
3 How to cure for Stretch the Use heating pad or Taking dosage of
cramp muscle which is warm water bag vitamin B
cramped and on tightened supplements to
tenderly rub it. muscles. get rid of cramps.

4 How to cure for If you are feeling Take the dosage Give a sponge Get plenty of
Cold and Fever? unwell, take a of aspirin. Contact bath with blood- rest, take
Paracetamol. a doctor if fever is warm water. steam.
longer than that
and doesn’t
respond to the
dosage.
5 I’m having This can be a Jaundice describes
symptoms of jaundice, You a yellowish tint to
yellow eyes and should consult a the skin as well as
nails? doctor. whiteness of eyes
turns yellow.
Table 3 :Review of result
Purpose Accuracy
Level

7 Proposed Model To provide 0.736

6 medical
suggestions,
5 symptom
4 checker and
introducing
3 expert system as
2
well.
[3] Health A Healthcare 0.616
1 Assistant Bot: A agent to provide
0 Personal Health medical
Assistant for the suggestion for
0.923 0.8 0.9 0.86 0.2 Italian Italian language
Language,2020 only.
Terms Present Terms Detected [7] An intelligent Providing 0.715
Chatbot using healthcare
deep learning suggestions
with BRNN using deep
Term Detection Ratio= Term Detected / Term Present ,2020 learning with
Bidirectional
RNN.
We noticed that with an average 73.6% of the non-
related terms present in our taken sample, our
Chatbot
Table 2 : Comparison with existing model
Executed terminology detection with 56.6% of
accuracy.

Post Answers Page :

Name of Expert : Nikita

Question : For how many days should I wait if I have a fever ?

Answer : You can wait upto 3 to 4 days. Still if you have fever longer than this, Consult a doctor.

Table 3 : Expert Answering Page

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
 VI. LIMITATIONS AND [3] Marco Polignano, Fedelucio Narducci, Andrea
FUTURE WORK Iovine, Cataldo Musto, Marco De Gemmis, and
Giovanni Semeraro, “A Personal Health Assistant
In this paper, we presented the healthcare chatbot for the Italian Language” 2020, IEEE.
which will help people to find solution for their
health problems at home. Currently we are working [4] Aleksandra Przegalinska, Leon Ciechanowski,
on only primary diseases but sooner we will Anna Stroz, Peter Gloor, Grzegorz Mazurek, Kelley
upgrade it with major diseases. We will try to make School of Business, Indiana University “In bot, we
healthcare chatbot more accurate. More features trust A new methodology of chatbot performance
will be added such as booking appointments, measures” 2019, Elsevier, ScienceDirect.
digital prescription, connecting with a consultant,
accessing all reports. [5] Ashwini Shangrapawar, Ankita Ravekar, Sakshi
Kale, Nidhi Kumari, Aman Shende, PankajTaklikar,
VII. CONCLUSION “Artificial Intelligence based Healthcare Chatbot
System” International Research Journal of
Engineering and Technology, e-ISSN: 2395-0056,
In this paper, we presented Healthcare Chatbot that p-ISSN: 2395-0072, Volume: 07 Issue: 02/Feb 2020
provides medical help to the patients for some of the
common diseases. The chatbot offers user the [6] Lekha Athota, Vinod Kumar Shukla, Nitin
medical help in the absence of doctor which will Pandey, Ajay Rana, “Chatbot for Healthcare System
help to improve the productivity & availability of Using Artificial Intelligence” 2020 8th International
the medical sector. The conversation between the Conference on Reliability, Infocom Technologies
system and user is performed through a chatbot and Optimization Amity University.
interface which is a simple way of interaction. There
are lots of models of medical chatbots out there [7] Manyu Dhyani, Rajiv Kumar G. L. Bajaj
which are comparatively high cost for a common Institute of Technology and Management, Greater
people and we are trying to control this kind of Noida, Uttar Pradesh, India “An intelligent Chatbot
inconvenience in our Healthcare Chatbot System. using deep learning with Bidirectional RNN and
This is actually a kind of symptom checker system, attention model” Elsevier, 16 May 2020.
which recognizes the patient's disease with 76% of
accuracy. Many models of medical chatbots out [8] Rhio Sutoyoa, Andry Chowandaa,, Agnes
there which are quite unaffordable for a general Kurniatia, Rini Wongsoa “Designing an
person but we are trying to overcome this limitation. Emotionally Realistic Chatbot Framework to
After the successful implementation of the system Enhance Its Believability with AIML and
we will work on appending more features or we can Information States” Elsevier, ScienceDirect, 19 Sept
say functions to the chatbot using appropriate 2019.
security. Additional functions might be booking
appointments, digital prescription, connecting with [9] M. Dahiya, Dept. of Computer Science,
a consultant, accessing all reports and so on. Maharaja Surajmal Institute, Janakpuri, India, “A
Healthcare chatbots have potential to improve the Tool of Conversation: Chatbot” JCSE International
healthcare sector. Journal of Computer Sciences and Engineering, 30
May 2017
VIII. REFERENCES
[10] Bhagyashree Mohanta, Dr. Srikanta Patnaik,
[1] Kavitha B. R., Dr. Chethana R. Murthy, Department of Computer Science & Engineering
“Chatbot for Healthcare System using Artificial S’O’A University Bhubaneswar, India, Dr. Priti Das
Intelligence” 2019, IJARIIT. Department of Pharmacology SCB Medical College
and Hospital Cuttack, India, “paradigm shift in
[2] Dinesh Kalla, Fnu Samaah, “Chatbot for Medical digital healthcare system using Artificial
Treatment using NLTK Lib”, IOSR Journal of Intelligence, IOT and 5G Communication” 2019
Computer Engineering (Jan - Feb 2020). (ICAML).

[11] Bali, M., Mohanty, S., Chatterjee, S., Sarma,

M., Puravankara, R.: Diabot: A predictive medical
chatbot using ensemble learning
[14] N-gram Accuracy Analysis in the Method of
[12] M. B. Hoy, Alexa, siri, cortana, and more: An Chatbot Response,
introduction to voice International Journal of Engineering &
assistants, Jan. 2018. Technology. (2018)

[13] S. Devi, Z. A. M. Merchant, M. S. Siddiqui, [15] Mrs Rashmi Dharwadkar1, [Link]. Neeta A.
and M. Lobo, ``Arti_cial Deshpande, A Medical
intelligence based personal assistant,'' Asian J. ChatBot, International Journal of Computer Trends
Converg. Technol., vol. 5, and Technology
no. 1, pp. 1_4, 2019. (IJCTT) – Volume 60 Issue 1- June 2018

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.
[16] Flora Amato, Stefano Marrone, “Chatbots
meet eHealth:
automat zing healthcare, May-2018.

[17] Bayu Setiaji, Ferry Wahyu Wibowo, "Chatbot

Human-to- Machine Conversation Modeling",
Intelligent Systems Modelling
and Simulation (ISMS) 2016 7th International
Conference.

[18] Aishwarya Hajare, Priyanka Bhosale, Rasika

Nanaware, Guruswami Hiremath, Chatbot for
education system ISSN: 2454-132X, Impact factor:
4.295, Volume 4, Issue 3, IJARIIT.

[19] D. Gujar, R. Biyani, T. Bramhane, S. Bhosale,

and T. P. Vaidya, ``Disease
prediction and doctor recommendation system,''
Int. Res. J. Eng. Technol.,
Tamilnadu, India, vol. 5, no. 3, pp. 3207_3209,
Mar. 2018.

[20] V.K Shukla and A Verma, "Enhancing LMS

Experience through AIML Base and Retrieval Base
Chatbot using R Language", 2019 International
Conference on Automation Computational and
Technology Management (ICACTM).

[21] Chin-uan Huang, Ming-Chin Yang, Chin-Yu

Huang, “A chatbot-supported Smart Wireless
Interactive Healthcare System for Weight Control
and Health Promotion”, proceeding of the IEEE,
April-2018.

[22] C.P. Shabariram, V. Srinath, C.S. Indhuja,

Vidhya (2017). Ratatta: Chatbot Application Using
Expert System, IJAR in Computer Science and
Software Engineering,2017

Authorized licensed use limited to: Somaiya University. Downloaded on October 12,2024 at 06:33:17 UTC from IEEE Xplore. Restrictions apply.