A Hybrid Approach to Develop and Integrate Chatbot in Health

Informatics Systems

Abhijat Chaturvedi, Siddharth Srivastava, Astha Rai and A. S. Cheema

Centre for Development of Advanced Computing, Noida, India

Keywords:

Chatbot, Health Informatics Systems.

Abstract:

In this paper, we develop a chatbot that seeks free-form natural language queries by its users for blood and

related services such as list of blood banks, live blood stock, blood donation camps etc. with one or more

parameters as search criteria. The queries can be both Frequently Asked Questions (FAQs) and data driven

including location based services. The uniqueness of this chatbot lies in the fact that its architecture provides it

ﬂexibility to evolve to encompass more domains and services without having any impact on existing services.

Moreover, with approximate keyword initialization, the proposed chatbot can smartly infer from incomplete or

incorrect queries by the users as well as has the ability to learn abbreviations. The bot achieves this by leverag-

ing state-of-the-art deep learning and natural language understanding algorithms at the back-end. Speciﬁcally,

this bot uses a hierarchical approach for parsing queries. At ﬁrst level, it parses the query into intents i.e. FAQ

or data driven. If the classiﬁed intent is a FAQ, chatbot to respond while, if it is amongst many of the citizen

centric queries, it drills down through the query to identify the entities such as city etc. along with the type of

the service and returns the users with the required details.

1 INTRODUCTION

As the reachability of various health informatics so-

lutions are increasing, the usability and effectiveness

of such systems have become important. While user

interfaces and workﬂow play an important role in ac-

ceptance of such systems, but letting the users query

the system with natural text as input and giving them

the appropriate output such as data, help text, ex-

cerpts from user manuals, ﬁltering their query for

more inputs, is becoming the next level of expertise

required in health informatics solutions. The chat-

bots are a natural solution to this problem (Chung

and Park, 2019). Chatbots brings the ease of use

along with 24X7 service at a fraction of what a hu-

man employee would require. This enhances the op-

erational efﬁciency at a reduced cost. As argued by

Petter Bae Brandtzaeg in his study(Brandtzaeg and

Følstad, 2017) people prefer chatbots because of pro-

ductivity. Chatbots in health care provides a variety

of solutions from symptoms checking (Divya et al.,

2018) to getting basic health information (James and

Vales, 2009). The AI provides the edge to the chatbots

over any other technology, because they can learn and

grow over time. This saves the time and money of

both patients and health care provider. Thereby, cre-

ating a smooth health care experience.

However, most of the chatbot solutions work on

static data or pre-deﬁned workﬂows which may cater

to speciﬁc requirements. Therefore, in this work,

we propose a modular chatbot framework for re-

quirements suiting to health informatics solution. To

demonstrate the effectiveness of the framework, we

take a ”Blood Bank Management System” as a use

case. Speciﬁcally, the considered blood bank man-

agement system has two modes of usage. The ﬁrst

is via a citizen centric portal where details on blood

banks, blood stock, information etc. are given. The

second is the set of users using the application for

day to day use such as blood bank staff. Therefore,

we implement the proposed framework for address-

ing queries from citizens via a nationwide public web

portal as well as the users of a blood bank manage-

ment system.

Figure 1 shows the workﬂow of queries from a cit-

izen centric portal for a blood bank within the pro-

posed architecture. It is important to note that, the

queries from both sets of users (citizens, application

users) are different. While we can limit the scope of

queries from internal users of the health informatics

solutions such as blood bank staff in the considered

use case, it is extremely difﬁcult and impractical to

control or limit the queries from citizens on a public

774

Chaturvedi, A., Srivastava, S., Rai, A. and Cheema, A.

A Hybrid Approach to Develop and Integrate Chatbot in Health Informatics Systems.

DOI: 10.5220/0009188007740781

In Proceedings of the 13th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2020) - Volume 5: HEALTHINF, pages 774-781

ISBN: 978-989-758-398-8; ISSN: 2184-4305

portal.

In view of the above, the contributions of this pa-

per are:

1. A generic chatbot framework for health informat-

ics solutions and implementation in a nationwide

blood bank management system.

2. Formalizing common problems and their solu-

tions while implementing chatbots.

The rest of the paper is organized as follows, in

Section 2 we discuss and compare various types of

chatbots. In Section 3, we formally deﬁne the prob-

lem, the challenges and relevant works. Section 4 de-

scribes the proposed architecture while Section 5 dis-

cusses the results. In Section 6 we discuss possible

extensions of the proposed architecture followed by

conclusions in Section 7.

2 TYPES OF CHATBOTS

On the basis of workﬂow, chatbots can be classiﬁed

into three categories as follows

• Menu based Chatbots: Menu based chatbots

work on the principle of decision trees. The hier-

archies are used to get the next stage of input from

the user and thereby classify the next best action

based on the input received from the user. These

chatbots full ﬁll the basic search of the queries

but the advance search with multiple variables can

not be handled from such type of chatbots. Also,

the menu based chatbots fall short on the number

of clicks required for getting the desired informa-

tion from the bot. It requires maximum number

of ﬂows to get the desired information among the

three types of chatbots discussed in this paper.

• Keyword based Chatbots: These types of chat-

bots works on the identiﬁcation of keywords.

Once the input is received from the user, it tries

to map with the existing set of keywords and once

a match is found the corresponding logic is exe-

cuted. This type of chatbots gives an illusion of

understanding the user. However, it just identiﬁes

the keywords in the user’s statement. Keyword

based chatbots fails when there is multiplicity in

the keywords and one keyword falls under differ-

ent intents.

Like in the case of blood stock search and blood

donation camp search. In both the intents, key-

words are same such as blood and search. Sim-

ilarly, the FAQ’s about blood donation also have

many common keywords.

• Contextual based Chatbots: Contextual chat-

bots understands the context of the conversation

and therefore are the most advanced form of chat-

bots. These bots utilize Machine Learning (ML)

and Artiﬁcial Intelligence (AI) to remember con-

versations with speciﬁc users to learn and grow

over time(Xiaojiang, 2014). These chatbots ﬁlls

the slots once the information is available and then

uses the same slots to understands the context of

the conversation. This helps the user because the

bot understands the history and ﬂow of the con-

versation and remembers the information it has

received from the user.

A common example of ﬂow of the contextual

chatbot is when user asks the availability of B pos-

itive blood in Delhi. The bot replies the stock list

of blood banks. Then the user inputs and in mum-

bai?. The bot understands that the user is asking

for availability of B positive blood.

3 BACKGROUND

3.1 Problem

While the paper aims at developing chatbots for

health informatics solutions and the proposed archi-

tecture is applicable to development of generic archi-

tectures, for clarity of the readers, we limit the scope

of the discussion for next two sections w.r.t. a blood

bank management system. Blood search, traditionally

was done over the web in the drop down select type

method, which is convenient but not enough as typing

a random query over the app. The same goes with the

frequently asked questions. Users do not want to read

all the question, if they want to know the answer of

just one. Therefore, to reduce the time, effort and ease

the complexity for users, a system is needed which

can be interactive and provide relevant information by

understanding the language of the user, who may or

may not be accustomed to the system. Therefore, we

propose to use chatbot for interacting with the users

and solving their queries for getting information or

providing them help while using the system.

3.2 Challenges

One of the biggest challenge in the process of cre-

ating the bot was the understanding of the user’s in-

tention in the different sounding statements. For in-

stance, the statements, ”Look for blood in Delhi” and

”I am looking blood banks in Delhi, India”, may have

different selection of words, but the intention of the

A Hybrid Approach to Develop and Integrate Chatbot in Health Informatics Systems

775

Figure 1: Workﬂow of the proposed framework.

user in both these sentences is the same. In both the

cases the user wants to search the blood for donation

in Delhi.

The correct mapping of the questions in the FAQ

section was another major challenge. In practice, in

any application, initially the FAQs are limited and

they grow as the application is developed over the

years and also with the number of users. Therefore,

with limited amount of questions in the beginning, it

is difﬁcult to correctly relate the user’s query to the

nearest question available in the dataset. For exam-

ple, when the questions in the dataset can be sim-

ilar (”Why should I donate blood?”, ”When to do-

nate blood?”) and the input from users can be lim-

ited (”Why donate blood?”), the algorithms usually

provide incorrect output as most of the sentences in

the dataset contain the word blood and when, why

(”When to donate blood instead of plasma?”).

3.3 Relevant Works

There has been no attempt at making a blood donation

search chatbot. Although there are many FAQ chat-

bots as well as many chatbots are tailored for health-

care industry, most of them are for disease prediction

like Mandy (Ni et al., 2017) and Quro (Ghosh et al.,

2018) or act as a counselling service (Lee et al., 2017).

The majority of these FAQ chatbots deploy a bag-of-

words model to predict the answer of the question en-

tered by the user. This technique, is however, not al-

ways effective especially in the cases where the length

of the questions is comparatively large.

3.4 Motivation

Ability to get information on availability of is a task

of utmost urgency. We wanted to provide a way that

is fast, easy to use and reliable when it comes to the

question of life and death. Hence we choose to create

a chatbot because that is the most easy way for a non

technical person to search for blood when required.

Also, every business has its own challenges and

requirements, it is not possible to ﬁt one solution to

every problem. Therefore, we wanted to provide a so-

lution for the better engagement of the users on the

business, through chatbots. Considering the major-

ity of the people who do not understand the complex

of higher machine learning, artiﬁcial intelligence and

natural language processing, we created a modular

approach to create a chatbot, which can be used as-

is or extended by adding, removing or modifying the

existing modules.

4 METHODOLOGY

The chatbot can is made up of two parts, one being the

FAQ section which handles the FAQs and the other be-

ing the custom section for all the other queries of the

user. In all total this bot handles 13 different intents.

The architecture of the bot is displayed in Figure 2.

HEALTHINF 2020 - 13th International Conference on Health Informatics

776

Figure 2: Architecture.

4.1 Intent Parsing

The ﬁrst step that is made in the process of chat-

bot providing an answer to the user’s query, once the

user’s input is captured is the intent classiﬁcation or

intent identiﬁcation. For this, Snips NLU is deployed.

Snips reads the text and based on it’s Deterministic In-

tent Parser(DIP) and Probabilistic Intent Parser(PIP)

classiﬁes the intent and the corresponding entities as

explained by Coucke et al. in 2018 (Coucke et al.,

2018). The step is vital since the output of this step is

the intent or the expectation of the user from the bot.

This output is the input to the next step in the process.

The Snips algorithm reads the text entered and

tries to map the statement with the statements pro-

vided in the training set. This is done by the com-

ponent called Deterministic Intent Parser(DIP). This

is swift and accurate segment of the Snips algorithm.

However, in most cases it is not possible to get the ex-

act statement as provided during the training phase.

Therefore, the second segment Probabilistic Intent

Parser(PIP) is put into place. If the DIP fails to pre-

dict the intent and entities PIP is initiated and on the

basis of probability model, it identiﬁes the intent and

entities of the user’s statement. See this structure in

ﬁg 2.

Once the intent is classiﬁed, next comes the task

of generating the response corresponding to the user’s

query. This is done through a series of custom pro-

cessing, business logic and api calls.

4.2 FAQ Section

To overcome the limitation of scarcity of data avail-

able to train, we used two fold mechanism. One is

to generate similar sounding questions from all the

questions in the dataset. This helped a lot in identi-

fying the correct question and thereby correct corre-

sponding answer that the user wants. Second was to

put the bot to use by actual users. We therefore pre-

sented some different users who used and queried the

bot just like an actual users. Each of them used the bot

in all intents and all use cases. Further processing the

FAQ’s multiple options were available, including ex-

ternal libraries and internal machine learning model.

In order to keep it simple and easy to use, we prefer

pre-build libraries rather than creating a ML model

from scratch. For this, many options were available

(Janarthanam, 2017), including DialogFlow and MS

Azure.

As studied by Canonico et al. in 2018 (Canonico

and De Russis, 2018), DialogFlow is the best avail-

able cloud based chatbot engine. It has high usability,

pre-build entities and pre-build intents. Also, it has

Default Fallback intent, which comes handy when an

unexpected statement is encountered.

The setup of DialogFlow started with the creation

of a Google account and creating an agent that will

process the information and produce the result. Once

done, the FAQ’s has to be uploaded in the csv format.

The DialogFlow, automatically learns the intents and

entities and trains the agent.

4.3 Custom Processing

For the search of blood along with the quantity and

camp search, the custom processing section is pro-

vided. This section basically handles the queries of

the blood search. Once the intent is classiﬁed the cor-

responding code gets invoked and the API calls are

made to the e-Rakt Kosh Web Services.

The user’s query contains the intent and the enti-

A Hybrid Approach to Develop and Integrate Chatbot in Health Informatics Systems

777

Figure 3: Chatbot framework based on Snips NLU architecture.

ties for the blood search. These intent and entities are

identiﬁed by the Snips algorithm and upon identiﬁca-

tion the API of the e-Rakt Kosh web services are in-

voked, the API returns the data in json format, which

is again parsed by the custom processing segment in

order to present the desired data in the desired format.

Intents like mobile apps for download and help are

also included for the better user aid. Quantity bound

blood search is an important part where most user en-

gagement is expected. Intent to contact the service

provider is also put into place so as to give the user

a smooth experience while chatting. This way users

need to put a single statement for getting the quantity

of blood available in a particular blood bank.

4.4 Analytics

MD Mulvenna et al. presented a way for analysing

the chat logs (Mulvenna et al., ). For the sake of fu-

ture development and to understand the usage pattern

of the users, we have deployed the analytic mecha-

nism for the bot. Through this we want to capture the

insights such as the most searched question or which

geographical area is using the bot most. This will help

us in creating a better user experiences by enhancing

the search and service accordingly.

The chat logs are captured in the SQL database,

from where it can be accessed for the analysis and

usage pattern understanding. Actionable analytics are

of immense help in expanding the services and user

experience in any business. The bot usage insights

will result in the better and smooth user experience,

which in turn enhance the user engagement.

4.5 Workﬂow

For instance the user enter the query, ”Look for ab+ve

blood in delhi”. The very ﬁrst step in the process will

be the parsing of the sentence in the Snips algorithm

which will classify the intent of the user, which in this

case is, Stock Search. The entities will be identiﬁed

as ”A+ve” and ”Delhi”. As they both are necessary

for the fetching of the list of blood banks.

Once the intent is identiﬁed, it will lead to the cor-

responding segment to be initiated to generate the cor-

responding answer. Here, it is important to note that

chatbot handles only one intent at a time. This allows

the bot to be simple and light weight.

Since the intent here is Stock Search, therefore

the bot will try to get the answer for that. The blood

group is identiﬁed as ”A+ve” and the city as ”Delhi”.

Now, the bot will get the geocodes of the said loca-

tion from Google Maps API. The geocodes are then

sent to the e-Rakt Kosh web service, which returns

the blood banks in the area.

Similarly, in case of fetching the list of blood

banks along with the minimum required quantity of

blood. The work-ﬂow remains the same except after

fetching the list, the bot ﬁlters out the blood banks

which have less than required quantity of blood units.

See workﬂow in ﬁg 3.

On the similar lines, if the statement is like ”how

much o+ve blood is avaiable in fortis hospital ,

noida”, the entities that will be extracted will be ”For-

tis hospital” as the name of the blood bank, ”O+ve” as

blood group and ”noida” as city. However the intent

will be count search instead of simply stock search.

Some other less used intents are also put in places

for one stop solution for the users. These includes,

a help intent which guides the user to use the bot in

HEALTHINF 2020 - 13th International Conference on Health Informatics

778

Figure 4: Workﬂow.

easy steps. Notiﬁcation intent is also present to fetch

any latest notiﬁcation about the blood bank portal.

4.6 Chat Logs

Every interaction with the bot is logged in the e-Rakt

Kosh database. This data is used for analysis of the

user engagement and enhancing the bot’s quality. The

questions asked by the user are directly put into ta-

bles. Corresponding to which intent as classiﬁed by

the bot is put in the table along with the response from

the bot. Once the chat is started the logging process

begins.

4.7 Integration

Once developed the bot needs to be integrated with

the erkatkosh website. This was the biggest challenge

of the project. The website is developed in java and

the bot is developed in python. Integration of the both

causes technical glitches and inefﬁciency.

Two clear options were available for the integra-

tion of the code. One was to compile the python code

in java and generate a java byte code. This is however

not a suitable mechanism because of its resource in-

tensiveness. The other method was to use api’s. This

way it will be easy to maintain and less resource in-

tensive. The most common to use python server is

ﬂask(Vogel et al., 2017)(Lokhande et al., 2015) which

hosts the application. Therefore we had to create an

instance of ﬂask and use it’s api’s to get the response

to the main page in e-Rakt Kosh website.

At the front end, there is the html, javascript and

css page for e-Rakt Kosh website with a popup of the

chatbot. This captures the users’s statement in the

textarea. Following it is the api call that sends this

statement to the ﬂask application running on another

port on the same server through post method. The

ﬂask application is the bot application itself. It re-

ceives the input processes it through business logic

and machine learning algorithm it has and sends back

Figure 5: Flask API Flow.

the generated response. This structure is shown in ﬁg-

ure 4. Flask API structure is shown in ﬁgure 5.

The simple and easy to use features are the key

components of the bot. This is to ensure that users

with all technical literacy are able to use the bot. Min-

imum button layout is preferred so as to clear the clut-

ter and give the user a smooth experience.

5 RESULTS

In 1950, Alan Turing proposes the Turing test for

the intelligent machines (Turing, 1950). Much later

in 2016, Zhou Yu et al. proposes the crowd source

method of evaluating the chatbot (Yu et al., 2016).

They propose a method of taking user feedback on the

chatbot usage. They argued that with modern chat-

bots, which are much more capable and there is no

end to the chat(in general chatbots) the Turing test

is not enough for performance evaluation. Marking

the chat as satisfactory or not, and whether the user

is willing to use the service again, provides a mecha-

nism to evaluate the chatbot.

We have used the ﬁrst method in our result ﬁxa-

tion. A dataset of random 206 questions from all in-

tents is used along with some questions that does not

fall in any intent. Such a dataset is put to test the in-

tent classiﬁcation of the bot. Out of the 206 random

questions, the bot correctly answered 179 questions,

while 26 questions were not identiﬁed by the bot. This

actually provides a near 87 percent accuracy.

However merely intent classiﬁcation of the user’s

statement is not enough for the bot. Generating the

A Hybrid Approach to Develop and Integrate Chatbot in Health Informatics Systems

779

correct statement for response is vitally important. In

order to check the correct response generation, the

same dataset is used to test the accuracy of the bot.

Out of 206 questions, the bot accurately answered 175

question working at an approx accuracy of 85 percent.

The dataset is entered in the database and a script

is used to fetch the question from this testing dataset

and putting it into the bat and ﬁnally saving the bot’s

response back in the database. This created a table

with a set of random question and answers on the ba-

sis of the chat session.

The second technique proposed by Zhou Yu et al.

is based on the capturing of the user feedback. This

evaluation technique is however, more subjective and

user centric, therefore the results vary to the great ex-

tent. They propose a system where the users are able

to rate the bot after a set of interaction with the chat-

bot.

Table 1: Intent Classiﬁcation and Response generation re-

sults.

Results

Measure Correctly

Identiﬁed

Incorrectly

Identiﬁed

Accuracy

Intent Classi-

ﬁcation

179 27 86.89

Response

Generation

175 31 84.95

6 GUIDELINES AND FUTURE

DIRECTIONS

In this section, we provide a few suggestions for repli-

cating and extending our work which will be of im-

portance for complete integration of chatbots in health

informatics systems.

• Analysis: The formal analysis of chatbot solu-

tions in an indistrial setting is important but is

given little attention. We recommend to setup an

R based analytic tool for the proper analysis of

the chat logs for both exploratory data analysis

and explanatory data analysis. This will provide

the business intelligence tools and graphs that are

easy to present and read information from. The

analysis also reveals the usage patterns and most

queried statements. This type of data gathering is

important for the better customization of the prod-

uct. It will also be helpful in identifying the gap

in the supply and demand of the blood in various

parts of the country.

• Voice based Chat: The chatbots can be extended

with TTS(Text to Speech) and STT(Speech to

Text), making it further easy for anyone to use

it. The voice capturing can be of immense help

when being used in mobile devices, thereby tap-

ping a huge market. There has been some of the

works in this area as well. Works of Tsiao et

al(Tsiao et al., 2007) and Emerick et al(Emerick

et al., 2015) notable in this area. The voice based

search combining with the regional language sup-

port will open this product to virtually everyone

holding a smartphone. In the context of this work,

this will let users search for blood in emergency

situations, irrespective of their technical literacy

and proﬁciency.

• Language: Efforts should be made to include re-

gional languages along with English, so as to en-

able every person to use this service, thereby re-

moving the language barrier. V

ıra et al (V

ıra et al.,

2014) demonstrated this with Spanish, French,

and Russian apart from English.

• Conversational Responses: The chatbot re-

sponds to the user messages if it is well trained on

the possible ﬂow of the conversation. This gives

the user an illusion that the bot is actually talk-

ing to him. However, the bot is just following the

pre-trained story line with different branches as

deﬁned by the developer team.

Other libraries such as Rasa NLU and Rasa Core

provides good enough structure to create a con-

versational chatbot(Bocklisch et al., 2017), that

keeps the track of the conversation between the

user and the bot and provides feature rich expe-

rience to the user. While we found that Rasa is

more ﬂexible, but it is slightly complicated (de-

pendencies and required data) for quickly getting

a system up and running as compared to Snips.

7 CONCLUSION

We have elaborated our proposed system in detail.

The proposed system will bring the ﬂexibility and

ease in the blood search at the national level. With the

use of Natural Language Processing, users can look

for blood in query format rather than looking up on

the web page. This saves time and effort which is

vital in cases of emergencies. The future plans also

includes the search in multiple languages which will

make the system more accessible for other languages

as well. The system will be the ﬁrst point of contact

between a potential donor and the e-Rakt Kosh appli-

cation. It will cater the FAQ’s to the donor and clear

the doubts regarding blood donation.

We have explained its integration with the exist-

HEALTHINF 2020 - 13th International Conference on Health Informatics

780

ing e-Rakt Kosh application and the challenges that

were faced including the different language and envi-

ronment

REFERENCES

Bocklisch, T., Faulkner, J., Pawlowski, N., and Nichol,

A. (2017). Rasa: Open source language under-

standing and dialogue management. arXiv preprint

arXiv:1712.05181.

Brandtzaeg, P. B. and Følstad, A. (2017). Why people use

chatbots. In International Conference on Internet Sci-

ence, pages 377–392. Springer.

Canonico, M. and De Russis, L. (2018). A comparison

and critique of natural language understanding tools.

CLOUD COMPUTING 2018, page 120.

Chung, K. and Park, R. C. (2019). Chatbot-based heathcare

service with a knowledge base for cloud computing.

Cluster Computing, 22(1):1925–1937.

Coucke, A., Saade, A., Ball, A., Bluche, T., Caulier, A.,

Leroy, D., Doumouro, C., Gisselbrecht, T., Calta-

girone, F., Lavril, T., et al. (2018). Snips voice plat-

form: an embedded spoken language understanding

system for private-by-design voice interfaces. arXiv

preprint arXiv:1805.10190.

Divya, S., Indumathi, V., Ishwarya, S., Priyasankari, M.,

and Devi, K. (2018). A self-diagnosis medical chatbot

using artiﬁcial intelligence. Journal of Web Develop-

ment and Web Designing, 3(1).

Emerick, C. T., Logan, J. R., Bates, R. A., and Wahl, J.

(2015). Voice assistant system. US Patent 9,171,543.

Ghosh, S., Bhatia, S., and Bhatia, A. (2018). Quro: Fa-

cilitating user symptom check using a personalised

chatbot-oriented dialogue system. Studies in health

technology and informatics, 252:51–56.

James, S. and Vales, I. (2009). Personal healthcare assis-

tant/companion in virtual world. In AAAI Fall Sympo-

sium Series.

Janarthanam, S. (2017). Hands-on chatbots and conver-

sational ui development: Build chatbots and voice

user interfaces with chatfuel, dialogﬂow, microsoft

bot framework, twilio, and alexa skills.

Lee, D., Oh, K.-J., and Choi, H.-J. (2017). The chat-

bot feels you-a counseling service using emotional re-

sponse generation. In Big Data and Smart Computing

(BigComp), 2017 IEEE International Conference on,

pages 437–440. IEEE.

Lokhande, P., Aslam, F., Hawa, N., Munir, J., and Gulam-

gaus, M. (2015). Efﬁcient way of web development

using python and ﬂask.

Mulvenna, M. D., Bond, R. R., Grigorash, A., O’Neill, S.,

and Ryan, A. Hilda-a health interaction log data anal-

ysis workﬂow to aid understanding of usage patterns

and behaviours.

Ni, L., Lu, C., Liu, N., and Liu, J. (2017). Mandy: To-

wards a smart primary care chatbot application. In

International Symposium on Knowledge and Systems

Sciences, pages 38–52. Springer.

Tsiao, J. C.-s., Chao, D. Y., and Tong, P. P. (2007). Natural-

language voice-activated personal assistant. US Patent

7,216,080.

Turing, A. (1950). Mind. Mind, 59(236):433–460.

ıra, I., Tesel¸skis, J., and Skadin¸a, I. (2014). Towards the

development of the multilingual multimodal virtual

agent. In International Conference on Natural Lan-

guage Processing, pages 470–477. Springer.

Vogel, P., Klooster, T., Andrikopoulos, V., and Lungu, M.

(2017). A low-effort analytics platform for visualizing

evolving ﬂask-based python web services. In 2017

IEEE Working Conference on Software Visualization

(VISSOFT), pages 109–113. IEEE.

Xiaojiang, D. (2014). Chatbot system and method with con-

textual input and output messages. US Patent App.

13/661,040.

Yu, Z., Xu, Z., Black, A. W., and Rudnicky, A. (2016).

Chatbot evaluation and database expansion via crowd-

sourcing. In Proceedings of the chatbot workshop of

LREC, volume 63, page 102.

A Hybrid Approach to Develop and Integrate Chatbot in Health Informatics Systems

781