DiabeticFoodBot: Food and Water Intake Recommender System for

Diabetics

Irwan Reza Firmansyah

, Z. K. A. Baizal

, and Ramanti Dharayani

School of Computing, Telkom University, Bandung, Indonesia

Keywords:

Food and Water Intake, System for Diabetics.

Abstract:

Diabetes is a non-communicable disease which is one of the highest causes of death in the world. Diabetics

need to arrange the schedule, amount and type of food and water consumed every day from a nutritionist to

regulate blood sugar levels so that complications do not occur. A recommender system for food and water

intake that has been validated by nutritionists is needed to assist diabetics in determining the nutrients con-

sume. In this study we develop Artiﬁcial Intelligence (AI) telegram chatbot called as DiabeticFoodBot. This

system can provide food recommendations and water intake for diabetics. There are many previous works that

developed recommender systems for diabetics. However, this study has not considered the amount of water

intake for diabetics. In addition, our research uses household size in presenting the results of recommendations

to make it easier for users to determine serving sizes without using a scale. We develop our system using on-

tologies with Semantic Web Rule Language (SWRL) because they are considered capable of providing better

performance. The DiabeticFoodBot validation result of 94.7 percent shows that our system can provide good

recommendation results for users.

1 INTRODUCTION

Diabetes is a non-communicable disease that causes

1.5 million deaths every year and affects 422 million

people worldwide. Most of people with diabetes live

in low to middle income countries (W.H.O., 2022).

In Indonesia, the highest mortality from chronic dis-

ease is caused by diabetes mellitus (Ofﬁce, 2022).

Diabetes is divided into four types, including IDDM

(Insuline Dependent Diabetes Mellitus) or type 1 di-

abetes caused by autoimmune Langerhans beta cell

damage, NIDDM (Non Insuline Dependent Diabetes

Mellitus) or type 2 diabetes caused by relative failure

of Langerhans beta cells and resistance insulin, gesta-

tional diabetes occurring during pregnancy, and spe-

ciﬁc diabetes (Rahman et al., 2018; Hardianto, 2020).

Diabetes can strike anyone regardless of age or race.

and more common in obese people. Diabetes can

cause complications in the heart, nerves, kidneys and

eyes (Ministry, 2013).

To help control the blood sugar and minimize

complications, diabetics must maintain the amount,

type, and schedule of meals every day (Jahidin, 2019).

Face-to-face consultation with a nutritionist is needed

to get food recommendations that is suitable for dia-

betics is not possible to do every day. Although now

telemedicine has developed which can help diabetics

get consultation with a nutritionist online, this is not

effective and efﬁcient if done every day and costs a lot

(Simatupang, 2020). Therefore, diabetics need to de-

velop a food recommender system that has been vali-

dated by a nutritionist and can provide recommenda-

tions automatically to manage their daily meal menu.

We found a lot of research that built a food rec-

ommender system for diabetics. However, these stud-

ies have not considered the water intake for diabet-

ics. Even though we found research in India that

applied hydrotherapy as a therapy to regulate blood

sugar for diabetics. Water is also a medium for de-

livering nutrients and other substances into the body’s

cells and removing toxic substances so that the water

consumed will affect the patient’s blood sugar (Lim

et al., 2011). Therefore, we are considering building

a recommender system that also regulates the amount

of water intake of users.

Ontology is used as a knowledge domain to rep-

resent various types of food and the amount of min-

eral water served for users. The Semantic Web Rule

Language (SWRL) search method is an option in in-

ducing food menus and mineral water intake to users

because it can provide more expressiveness compared

to using a relational database which cannot produce

314

Firmansyah, I., Baizal, Z. and Dharayani, R.

DiabeticFoodBot: Food and Water Intake Recommender System for Diabetics.

DOI: 10.5220/0012639000003848

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Advanced Information Scientiﬁc Development (ICAISD 2023), pages 314-320

ISBN: 978-989-758-678-1

output if the data sought is not found. Several stud-

ies have also used ontology and SWRL to recommend

food menus (Cahn, 2017).

There is study that uses conversational systems to

interact with users. However, the ﬂexibility in conver-

sional systems that use chatbots is better, so it needs

to be considered (Baizal et al., 2020). Chatbots have

the ability to interact with users without the need for

a third person to provide feedback by utilizing Arti-

ﬁcial Intelligence (AI). Chatbot through Telegram so-

cial media is used to make it easier for users to interact

with the system through applications that are world-

wide and familiar to users and can be downloaded for

free on their smartphones (Editor, 2019; Arjun and

Baizal, 2022; Nurani et al., 2022). Based on the prob-

lems discussed in this study, we developed Diabet-

icFoodBot using the Telegram Chatbot based ontol-

ogy and SWRL. This system was developed to make

it easier for diabetics to determine the amount, type

and time of food consumption based on the valida-

tion results of nutritionists. This recommender system

provides information regarding the amount of water

consumption required by the user. Our research also

uses matrix unit serving sizes and household measure-

ments in presenting recommendations to make it eas-

ier for users to determine food portions.

1.1 Related Work

Several researchers have conducted study related to

determining the diet for diabetics. The goal of their

study is to automatically provide dietary recommen-

dations for diabetics using ontology. At the testing

stage, the researchers collected data from diabetics.

The data obtained is used as a query object. The re-

sults displayed are in the form of a food menu ar-

rangement with units of calories or weight units in

grams (Mckensy-Sambola et al., 2021; Arwan et al.,

2013; Farman Ali et al., 2017; Rachman and Nurjan-

nah, 2019).

Research using ontology design has problems in

displaying results that must be in accordance with the

database. If the database does not contain the infor-

mation the user wants, then the recommender system

cannot display the results. This problem is answered

in a research article that optimizes the recommender

system application using SWRL. SWRL is still able

to provide the information the user wants if similar

results are not found in the database by reviewing the

available database similarities. The results of the ac-

curacy of these studies can still be improved (Baizal

et al., 2020; Nurani et al., 2022; Arwan et al., 2013).

The use of chatbots in a food recommender sys-

tem for diabetics has also been used in Thailand and

has received a positive response to the user experience

due to its attractive appearance, easy use, fast per-

formance, and clear content. However, this research

only uses the Google Sheet system as a database

(Thongyoo et al., 2020). The use of Google Sheets

as a database has drawbacks such as error-prone in

data input, slow loading speed, limited data types that

can be stored, limited number of record storage, less

structured, and less speed in searching for data (Weir

et al., 2010). The use of chatbots with SWRL anal-

ysis on telegrams is also used in ﬁlm recommender

systems, but does not apply ontologies (Nurani et al.,

2022).

Based on previous research studies, the develop-

ment of a food recommender system for diabetics can

be optimized by representing ontology and SWRL us-

ing a telegram chatbot. The addition of a mineral wa-

ter intake recommendation feature was also added in

this study to maximize the user’s blood sugar control.

1.2 Related Theory

The total daily calorie requirement is determined us-

ing the Basal Metabolic Rate (BMR) formula mul-

tiplied by the Activity Factor (AF) in daily activity.

BMR is the calories needed by the body to perform

basic functions (Sihombing, 2017; Harris and Bene-

dict, 1918).

Male BMR = 88, 362 + (13, 397weight [kg])+

(4, 799 x height [cm])−

(5, 677 x age[years]

(1)

Female BMR = 447, 593 + (9, 247weight [kg])+

((3, 098 x height [cm])−

(4, 330 x age [years]

(2)

The user’s daily caloric needs are calculated using

the Body Mass Index (BMI), Basal Metabolic Rate

(BMR), and input data regarding the user’s daily ac-

tivity intensity. BMI is used to interpret the user’s

weight status. While the user’s activity intensity is

used as a supporting factor to determine daily caloric

needs (Harris and Benedict, 1918).

BMI =

Mass(kg)

Height(m)

(3)

BMI interpretation Table is used to determine the

user’s weight status (Indonesia, 2018).

Users’ daily activity intensity Table 2 is little or

no exercise, light, moderate, and vigorous exercise

(Mawartika and Guntur, 2021).

DiabeticFoodBot: Food and Water Intake Recommender System for Diabetics

315

Table 1: Body mass index.

BMI Weight Status

< 18.5 Underweight

18.5-22.9 Normal (Ideal)

23-24.9 overweight (Overweight)

25-29,9 Obesity I

> 30 Obesity II

Table 2: Daily activity and activity Factor.

Daily Activity Activity Factor (AF)

Man Woman

Little or no exercise

(Only light daily activi-

ties)

1.30 1.30

Light exercise (1-3 days

a week or light daily ac-

tivities)

1.56 1.55

Moderate exercise (3-5

days a week or moderate

daily activity)

1.76 1.70

Vigorous exercise (6-7

days a week or doing

strenuous daily activi-

ties)

2,10 2.00

The recommended formula for daily calorie needs

uses:

Daily Calories = BMR x AF ± Calories BMI (4)

BMI calories for the normal level are increased by

0 calories, the underweight level is increased by 500

calories, the overweight level is reduced by 300 calo-

ries, and the obese levels 1 and 2 are reduced by 500

calories. This calculation is used to improve the ac-

curacy of the calculation based on the intensity of the

activities carried out and considering the user’s BMI

level.

The calculation of the mineral water intake needed

by the user is (Gunardi, 2022):

Water Intake =

Weight(kg)x30

1000

(5)

The unit used in this calculation is liters. Calcula-

tion of daily nutritional needs is obtained through the

formula Percent Daily Value.

%DV =

NutritiononFood(g)

Daily Value

x100 (6)

The author uses the FDA-approved 2022 Daily

value Table 3 to calculate the user’s daily requirement

(F.D.A., 2020).

Meal portions each day are divided into 25% of

daily calories for breakfast, 25% of daily calories for

Table 3: Daily value.

Nutrition Daily Values

Carbohydrate 300g

Proteins 50g

Fat 78g

lunch, 20% of daily calories for dinner, and 30% of

daily calories for snacks according to what has been

agreed upon by a nutritionist.

2 METHODOLOGY

2.1 System ﬂow

In this research, we developed a chatbot on the Tele-

gram platform that can be accessed by users via lap-

tops, mobile phones or computers. The system re-

ceives input from the user in the form of user informa-

tion such as height, weight, daily activity, age, gender,

and user’s allergy history. This information is sent

to the handler to be changed in the form of a query.

Based on this query, the system performs reasoning

on system knowledge (Ontology and SWRL Rules)

to be able to provide recommendations. The recom-

mendation results are sent to the user via the chatbot

interface. This ﬂow is shown in Figure 1.

Figure 1: Chatbot system ﬂow.

2.2 System development

2.2.1 Knowledge development (Ontology and

SWRL)

There are four main classes that will be applied to the

ontology design. The main class is BMI, User, Al-

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

316

lergy, and Menu. The main ontology shown in Figure

Figure 2: Chatbot system ﬂow.

BMI class hierarchy is classiﬁcation based on

calculated values from user input using height and

weight as parameters. The subclasses of the BMI

class are obesityI, obesityII, overweight, normal and

underweight Figure 3.

Figure 3: The hierarchy of BMI class.

The User class stores user characteristic informa-

tion. Meanwhile, the Menu Class stores information

regarding the types of drinks and food. The sub-

classes of the Menu class are Drink, Snack, and Food.

The Food subclass stores data about fruits, carbohy-

drates, proteins, and vegetables Figure 4.

Figure 4: The hierarchy of Menu class.

The snack subclass stores data about snack such

as chips and cake. The Allergy class will store the

various types of allergies considered such as, nuts, di-

ary, oats, chicken, and seafood. Rule development

is adjusted to the calculation of the necessary needs.

An example of this calculation is the determination of

BMI and BMR values.

2.2.2 Information Retrieval

Information retrieval is the initial stage to obtain in-

formation from users. This process is carried out by

requesting data input from the user in the form of

functional requirements that act as properties in the

class, including information about the user’s height,

weight, daily activity, age, gender, and allergy history,

which is stored in the User Class.

2.2.3 Knowledge Validation

The validation process in this study was carried out

by nutritionists. Nutritionists validate datasets for on-

tology development. They also validate SWRL.

3 RESULTS

3.1 Implementation of Ontology and

SWRL

We use prot

e version 5.5.0 for composing the on-

tology. We use the top-down (tree) technique which

is deﬁned by forming classes, continuing into sub-

classes and ending with instances. The hierarchy of

class is shown in Figure 5.

Figure 5: The hierarchy of class in ontology.

Each class in the ontology has a data property.

This Data Property serves to complete the information

of each class. In addition to data properties, object

properties are also deﬁned to link between instances

of each class through semantic relations. Figure 6

shows the data properties and object properties that

DiabeticFoodBot: Food and Water Intake Recommender System for Diabetics

317

are used to create hierarchies and conceptual relation-

ships between instances.

Figure 6: Data properties and object properties.

The User class has a data property:

• hasBMI: the property that store the BMI calcula-

tion result data.

User (?p)

asweight(?p, ?w)

asheight(?p, ?h)

wrlb :

mul tiply(?wh, ?w, 10000)

wrlb :

mul tiply(?hm, ?h, ?h)

wrlb :

divide(?bmi, ?wh, ?hm)− > hasBMI(?p, ?bmi)

BMI calculation will be obtained to classify user

level BMI. For example, if the user’s BMI is less

than 18.5, it is considered underweight.

User (?p)

asBMI(?p, ?bmi)

wrlb :

greaterT han(18.5, ?bmi)− > underweight(?p)

Recommendations are given to users in the form

of 5 menus (Figure 5), that are breakfast menu,

morning snack, lunch, afternoon snack and din-

ner. The process is built using object properties

as can be seen in ﬁgure 5. LevelNutrient is a

property used to see whether the food is included

in the appropriate nutritional category to be rec-

ommended to users based on the nutrients in the

food. The calculation is obtained based on the

daily value with the SWRL rule example as fol-

lows:

• High fat

Menu(? f )

asNutrientFat(? f , ?n)

wrlb :

divide(?h, ?n, 78)

wrlb :

mul tiply(?DV, ?h, 100)

wrlb :

greaterT hanOrEqual(?DV, 20)− >

LevelNutrient(? f , HighFats)

• Low Carb

Menu(? f )

asNutrientCarbo(? f , ?n)

wrlb :

divide(?h, ?n, 300)

wrlb :

mul tiply(?DV, ?h, 100)

wrlb :

greaterT hanOrEqual(5, ?DV )− >

LevelNutrient(? f , LowCarbo)

The Menu class has a data property:

• hasServingSize: the property that will deﬁne the

food in terms of serving size measures. Described

in string form.

• hasWater: property to store the recommended

amount of mineral water intake. The following is

the SWRL rule formula for hasWater:

User (?p)

asweight(?p, ?w)

wrlb :

mul tiply(?wt, ?w, 30)

wrlb :

divide(?water, ?wt, 1000)− >

hasWater(?p,?water)

3.2 DiabeticFoodBot Prototype

The recommender system provides a recommenda-

tion menu according to user information. The chatbot

development process is carried out using the Python

language. For database development and queries, we

use sqlite3. Furthermore, to connect the program to

Telegram, we use the provided Telegram API. The

system receives input in the form of user informa-

tion, that is gender, age, activity intensity, weight,

height, temporary blood sugar, diabetes symptoms,

diabetes complications, and allergies. The system rec-

ommends food menus and mineral water intake, in-

cluding information about nutrition and when to con-

sume food. The conversation ﬂow process is shown

in Figure 7.

Figure 7: The hierarchy of BMI class.

The language used in this chatbot is Indonesian.

An example of a chatbot conversation is shown in Fig-

ure 8.

Figure 8: Interaction between user and DiabeticFoodBot on

telegram.

3.3 DiabeticFoodBot Testing

There are two steps to the testing process, that is test-

ing during training and system testing. Testing dur-

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

318

ing the training process is carried out by testing the

phrases on the system and checking the response sys-

tem. If the answer is incorrect, then additional train-

ing testing is required. After the system development

is complete, system testing is carried out. Simulation

by trying to provide input in the form of user infor-

mation, for example by looking at whether the BMI

results are true or false based on user information. If

the testing system is appropriate, the researcher con-

tinues to enter the ﬁnal stage, that is testing the results

of recommendations to nutritionists.

4 EXPERIMENT

The testing process of the recommendations by Dia-

beticFoodBot involves nutritionists. The nutritionist

validates the food list from the results of our recom-

mender system in Spreadsheet. Nutritionist validation

results are used to obtain true positive, false positive,

and false negative values. Based on these values, F-

Score, precision, and recall calculations can be per-

formed to see the accuracy of the recommended re-

sults.

Sample user data for the validation process were

obtained from diabetics who ﬁlled out a Google form

with a minimum age limit of 12 years. The num-

ber of sample user data obtained is 60 samples, but

three sample cannot be included because the user is

a diabetic with complications. Therefore, the total

samples used for validation were 57 user samples and

produced 285 samples of food recommendations and

mineral water intake. Of the 285 food samples and

mineral water intake approved by nutritionists, there

were 31 food samples that were inappropriate.

Precision =

T P

T P + FP

285

285 + 31

= 0.901 (7)

Recall =

T P

T P + FN

285

285 + 0

= 1 (8)

Recall is the ability of the system to retrieve the

appropriate document, while precision is used to mea-

sure the effectiveness of a system to ﬁnd information.

If the recall and precision values are close to 1 then

the results are good.

Precision and Recall are used to obtain the F-

Score, which is the average value of precision and

recall. This value can be obtained by the following

equation:

F − Score = 2x

PrecisionxRecall

Precision + Recall

0.901x1

0.901 + 1

= 94.7%

(9)

Table 4: Information of confusion matrix.

Information

TP (True

Positive)

The total results of food rec-

ommendations that are in ac-

cordance with the recommenda-

tions of nutritionists

FP (False

Positive)

The total results of food rec-

ommendations that are recom-

mended by the system but not

recommended by nutritionists

FN (False

Nega-

tives)

The results of food recommen-

dations that are not recom-

mended by the system and not

recommended by nutritionists

F-1 Score illustrates the comparison of the average

precision and recall listed. The level of accuracy is

shown from the F-1 Score presentation which is close

to 100%.

5 CONCLUSIONS

DiabeticFoodBot is a chatbot that recommends food

menus and mineral water intake for diabetics without

complications. Based on the validation that has been

carried out, the results obtained that the accuracy level

of this chatbot is 94,7 percent so that this chatbot is

able to provide food and mineral water recommen-

dations according to the user’s nutritional needs and

can be used as a solution to assist users in implement-

ing good eating and drinking patterns for sufferer di-

abetes.

The limitation of this research is the limit on the

server and database which hinders the speed perfor-

mance in displaying output results to users. In ad-

dition, researchers have also not assessed user opin-

ions regarding the displayed chatbot features. This

chatbot also cannot be used for diabetics who have

certain complications because diabetics with compli-

cations have different food and mineral water intake

rules, depending on the type of complications they are

suffering from.

REFERENCES

Arjun, A. R. and Baizal, Z. (2022). Chatbot-based movie

recommender system with latent semantic analysis on

telegram platform using dialog ﬂow. Journal of Com-

puter System and Informatic, 3(4).

Arwan, A., Sidiq, M., Priyambadha, B., Kristianto, H., and

Sarno, R. (2013). Ontology and semantic matching for

diabetic food recommendations. Int. Conf. ICITEE,

pages 170–175,.

DiabeticFoodBot: Food and Water Intake Recommender System for Diabetics

319

Baizal, Z., Widyantoro, D., and Maulidevi, N. (2020). Com-

putational model for generating interactions in conver-

sational recommender system based on product func-

tional requirements. Elsevier Vol.128.

Cahn, J. (2017). CHATBOT : Architecture, Design, De-

velopment. Thesis University of Pennsylvania School

of Engineering and Applied Science Department of

Computer and Information Science, Apr.

Editor (2019). Bmr calculator.

https://www.diabetes.co.uk/bmr-calculator.html,

Accessed Apr 28, 2022.

Farman Ali, S. I., Kwak, D., Khan, P., Ullah, N., and Sang-

jo Yoo, K. (2017). Type-2 fuzzy ontology-aided rec-

ommender systems for iot-based healthcare. Elsevier,

pages 138–155,. .

F.D.A. (2020). Daily Value and Percent Daily Value :

Changes on the New Nutrition and Supplement Facts

Labels. Department of Health Human Services USA.

Gunardi, A. J. (2022). Must know, water needs can be

seen from body size, you know! Online Website,

page – – – – – – –. .https://www.klikdokter.com/info-

sehat/kesehatan-umum/wajib-tahu, accessed April 20,

2022.

Hardianto, D. (2020). A comprehensive review of diabetes

mellitus: Classiﬁcation, symptoms, diagnosis, preven-

tion, and treatment,”indonesian. Journal of Biology

Biosciences (JBBI, pages 304–317,.

Harris, J. and Benedict, F. G. (1918). A biometric study

of human basal metabolism. Biological Sciences,

4(12):370–373,.

Indonesia, K. K. R. (2018). Klasiﬁkasi obesitas setelah pen-

gukuran imt. Accessed Apr 28, 2022).

Jahidin, A. (2019). Therapeutic effect of drinking water on

reducing blood sugar levels (gds) in type ii diabetes

mellitus patients”. Journal of Health. Generation De-

velopment, 11(1):87,.

Lim, J., Chan, M. M., Alsagoff, F. Z., and Ha, D. (2011).

Innovations in Non-communicable Disease Manage-

ment in ASEAN : A Case Series. Taylor Francis On-

line.

Mawartika, Y. E. B. and Guntur, M. (2021). Application

expert system for food selection based on nutritional

needs using forward chaining. Cogito Smart Journal

Vol.7, (1):96–110,.

Mckensy-Sambola, D., Garcia-Sanchez, M. A. R.-G. F.,

and Valencia-Garcia, R. (2021). Ontology-based

nutritional recommender system. Appl. Sci Vol.12,

(1):143,.

Ministry, M. O. H. o. t. R. o. I. (2013). Di-

abetes mellitus is the number 6 cause

of death in the world. Online Website.

https://www.kemkes.go.id/article/view/2383/diabetes-

melitus-pembebab-kematian-nomor-6-di-dunia-

kemenkes-tawarkan-solusi-cerdik-melalui-

posbindu.html, Accessed March 26, 2022.

Nurani, S., Baizal, Z., and Ikhsan, N. (2022). Sellybot :

Conversational recommender system based on func-

tional requirements. In 2022 International Conference

on Data Science and Its Applications (ICoDSA, pages

315–319,. IEEE.

Ofﬁce, W. K. P. H. (2022). Diabetes causes the

highest mortality in indonesia, overcome it

quickly before it’s too late. Online Website.

https://dinkes.kalbarprov.go.id/diabetes-sebabkan-

kematian-tertinggi-di-indonesia-atas secepatnya-

before-late/.(accessed December 20, 2022.

Rachman, D. D. and Nurjannah, D. (2019). Ontology

model-based recommender system in providing food

diet recommendations for diabetics. Telkom Univer-

sity e-Proceeding of Engineering, 6(2):9569,.

Rahman, A. Z., Soesanto, I., and Wahyunggoro, O. (2018).

Review on the food recommender system for diabetes

mellitus. IEEE. int. Conf. ISRITI, pages 627–632,.

Sihombing, M. (2017). Factors associated with hyper-

tension among diabetes mellitus people of indonesia

(basic health research 2013,”indonesian. Bulletin of

Health Research, 45(1):53–64,.

Simatupang, R. (2020). Dietary guidelines for people with

diabetes mellitus.

Thongyoo, P., Anantapanya, P., Jamsri, P., and Chotipant,

S. (2020). A useralized food recommendation chatbot

system for diabetes patients. Int. Conf. on Cooperative

Design, Visualization and Engineering, pages 19–28,.

Weir, R., Timms, G. P., Smith, K. A., Koury, R., and Pan, D.

(2010). Innovative Practices in Electronic Resources

and Acquisition Management. Purdue University.

W.H.O. (2022). Diabetes. https://www.who.int/news-

room/fact-sheets/detail/diabetes, accessed December

20, 2022.

ICAISD 2023 - International Conference on Advanced Information Scientiﬁc Development

320