Ontology Modelling of Malaysian Food Exchange List

Norlia Yusof

, Shahrul Azman Noah

and Samirah Taufiq Wahid

Knowledge Technology Group, Center of Artificial Intelligence Technology,

Faculty of Information Science and Technology, Universiti Kebangsaan Malaysia, 43600 Bangi Selangor, Malaysia

School of Healthcare Sciences, Faculty of Health Sciences, Universiti Kebangsaan Malaysia,

Jalan Raja Tun Muda Abdul Aziz, 50300 Kuala Lumpur, Malaysia

Keywords: Design, Case-based Reasoning, Case Adaptation, Constraints, Ontology, Dietary Menu Planning.

Abstract: Designing dietary menu planning is a complex problem-solving task. It involves several constraints and

extensive common sense. Case-based reasoning (CBR) solves the complexity by storing an expert common

sense in the case base. Case adaptation (CA) is important for design task using CBR since old cases are

partially similar as a current one. An automatic CA mainly focused on the processing level rather than at the

data level. On the other hand, semantic technology (ST) inserts the intelligence features by shifting the

focus on the application code to the data. This can leverage the burden on the logical processing of

adaptation engine. Ontology is a prerequisite in ST. Thus, this research proposes a computational model of

design CA using an ontological approach. This paper discusses the experience we gained during the process

of ontology modelling based on the OD101 method. The Malaysian food ontology was successfully

developed to make the domain assumptions explicit in supporting the reasoning process of case adaptation

for dietary menu planning recommendation.

1 INTRODUCTION

Designing a dietary menu planning is a complex

process. A dietitian needs to incorporate several

constraints, including numeric nutrition constraints,

personal food preferences and aesthetic criteria. All

of these constraints have to simultaneously satisfy.

Marling et al., (1999) stated at least three factors that

make designing dietary menu planning difficult i.e.

interrelated and unconstructive constraints and

extensive of common sense. Common sense is vital

in order to generate the sensible menu for a patient.

Although few applications (Noah et al., 2004;

Saludin et al., 2010; Chien-Yeh et al., 2011) have

been developed to assist in dietary menu planning,

they are still lack the common sense elements. It is

difficult to design dietary menu planning from

scratch due to the great amount of the common

sense. Marling (1996) proposed Case-Based

Reasoning (CBR) as the solution to this problem. By

using CBR, the menu generated will never go

implausible because the expert’s common sense

were embedded in the case base.

Design is one of the synthesis tasks in CBR. It

involves the redesign process to generate a new

design solution. This process is essential because

existing design is rare to exactly match the demands

of a new requirement. In CBR, redesign is

performed by case adaptation. Case adaptation is the

most difficult of CBR’s cycle to be implemented

specifically when involving constraints. Extensive

application code programming is required for their

heuristics process in automatic adaptation. In other

words, efforts in automating the case adaptation

algorithm have been mainly focused at the

processing level rather than at the data level. Any

changes in the adaptation algorithm require

extensive changes to the developed programming

code.

The emergence of semantic technology (ST) can

leverage the burden on the processing level of

adaptation strategy since the intelligence aspect is

encoded in the data rather than embedded in the

application code. The reasoning process of

adaptation strategy executed by means of ontology

which is the main component in ST. Thus, this

research aims to propose a computational model

using an ontological approach for design case

adaptation.

In this paper, we discuss how to model and

Yusof, N., Noah, S. and Wahid, S..

Ontology Modelling of Malaysian Food Exchange List.

In Proceedings of the 7th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2015) - Volume 2: KEOD, pages 301-306

ISBN: 978-989-758-158-8

301

develop the ontology which supports the constraint-

based case adaptation of a domain. The domain in

concern is the dietary menu planning for diabetic

patients where various constraints must be met

before a suitable menu can be generated. To this

aim, the paper is structured as follows. The domain

of dietary menu planning for diabetic is concisely

described in section 2. Section 3 explains the

motivating scenarios of the application. Section 4

discussed the tasks executed according to the

ontology development methodology to build the

ontology. The experience gained from the ontology

development is discussed in section 5. And finally,

section 6 concludes the implementation of the

ontology.

2 DIETARY MENU PLANNING –

THE DOMAIN

Dietary menu planning focus on a patient where

therapeutic diet is one of the treatments that can

improve their health conditions. Diabetes is one of

the diseases where healthy eating is the key feature

in its treatment plan. Healthy eating can be achieved

through balanced intake of macronutrients i.e.

carbohydrate, protein and fat. A dietitian works out

with a diabetic to design a healthy personalized

dietary menu plan. This menu plan not only sounds

nourish but, it has to consider the personal food

preferences of a diabetic.

Foods are groups into eight main groups,

including starch, vegetables, fruits, milk and yogurt,

seafood, meat, plant-based proteins and fats. Each

food contains nutritional values i.e. energy,

macronutrients, vitamins and minerals such as

sodium and potassium. Foods are measured

according to its serving size. The unit of serving size

is refer using a household measures such as cup,

slice, tablespoon and teaspoon. The food groups, the

nutrient values and the serving size are referred as

general information of foods and also known as

nutrient data. In Malaysia, these nutrient data are

published in the food composition database namely

nutrient composition of Malaysian foods (Tee et al.,

1997). Another reputable source that provides

nutrient data is the Unites States Department of

Agriculture (USDA) nutrient database.

Other information that can be tagged on a food

such as does the food contains any one of eight

allergy foods i.e. milk, egg, peanut, tree nuts, fish,

shellfish, wheat, and soy, the suitable time to take

this food, is it during breakfast, lunch or dinner, is

this food has been allowed in a religion, which

cultural this food is belong, is this food a superfood

for a diabetic. This information is referred as the

auxiliary information to the food.

The nutrient data is used for numerical nutrition

constraints in dietary menu planning system. This

data is required to calculate the total of energy,

macronutrients, cholesterol, vitamins and minerals.

At this point, serving size plays very important role

to ensure the dietary menu planning do not exceed

the proportion of macronutrients that has been

allocated to a diabetic according to their calorie

needs. Nevertheless, serving size calculation is a

challenging task, and it has become one of the

reasons the urge of dietary menu planning

automation (Kovacic, 1995).

Warshaw (2010) stressed the importance of

personalization in healthy eating plan for a diabetic.

Personalization is related with food preferences,

including like and dislike foods, foods that cannot be

eaten due to the health matter or also known as

allergy foods, cultural and religious food habits.

Thus, food preferences become one of the

constraints needs to be fulfilled when designing a

dietary menu planning. The auxiliary knowledge of

foods is utilized to solve the personal food

preferences constraints.

3 MOTIVATING SCENARIOS

A scenario was outlined to show how significant the

ontology in solving the problems by providing

possible solutions. We designed a possible scenario

where ontology is used to support the reasoning

process of adaptation engine in dietary menu

planning application.

The main actor in our scenario is Miss Anna,

who is the registered dietitian. She is specializes in

diabetes medical nutrition therapy (MNT). Miss

Anna wants to make a personalized dietary menu

planning recommendation of her patient, John who

had type-2 diabetes. At first, she calculates John’s

calories level which is 1800 kcal. By using this data,

she is able to design the mealtime exchange table

(MET) as shown in Table 1. This table is used as

guideline to ensure the dietary menu planning is

balanced accordingly to the eight food groups for

each of the meal time. Then, she starts using the

dietary menu planning system to make the

recommendation based on the nutrition and personal

preference constraints of her patient.

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

302

Table 1: Meal exchange table of 1800 kcal.

Food group Exc BF MS L AS D

Starch 9 3 - 3 - 3

Vegetable 4 - - 2 - 2

Fruit 3 - 1 1 1 -

Milk 2 - 1 - 1 -

Fish 4 - - 4 - -

Meat 3 - - - - 3

Plant based 1 1 - - - -

Fat 8 2 - 3 1 2

Exchange (Exc), Breakfast (BF), Morning snack (MS), Lunch

(L), Afternoon snack (AS), Dinner (D).

She input the patient details such as calorie level,

body mass index (BMI) classification, religion and

race where these data are used as indexed features,

personal preferences i.e. like and dislike food,

allergy and prohibit food and MET. The system

retrieves the best case based on the similarity of the

indexed features of new case with the old case. The

best case will be adapted if it does not comply with

the constraints. The first constraint is MET. The

adaptation process checks whether each of the food

group in MET has been fulfilled by the best case. If

there is any food group in MET does not exist in the

best case, adaptation engine will make a suitable

recommendation of the food belong to the food

group. The next constraints are the forbidden foods.

The adaptation process checks does the best case

consist of any forbidden food. If the best case has

this kind of food, adaptation engine will substitute it

with an eligible food. Adaptation process also

checks the food accompaniment between foods that

has been suggested by the adaptation engine so that

the menu planning is sensible. Finally, it checks the

serving size of food items to ensure the meal

planning is within the calorie range. If the serving

size is less than the required value, adaptation engine

will make a recommendation of a new food to fulfil

the serving size. The output from system is the

recommendation of personalized dietary meal

planning for the patient.

The adaptation engine use ontology to perform

the semantic reasoning. A reasoner in the ontology

can compute the superclass-subclass relationship or

also known as class subsumption automatically. This

feature can be used in the adaptation engine to make

a recommendation of new food for MET and serving

size constraints. While forbidden food and food

accompaniment constraints exploit the reasoning

through the object property that assigned to each of

the food instances.

4 BUILDING THE MALAYSIAN

FOOD EXCHANGE LIST

ONTOLOGY

We applied the Ontology Development 101

(OD101) method proposed by Noy and Mcguinness

(2001) as the guideline to build our first ontology of

Malaysian foods exchange list (MyFELO).

4.1 Specification: Determine the

Domain, Purpose and Scope of the

Ontology

The first step in OD101 is specification where its

domain, purpose and scope were defined. The

representation of MyFELO is the domain of the

ontology. Two main purposes of building MyFELO

are to represent the general knowledge using in

hierarchical form and to reason the general

knowledge in supporting the constraint-based case

adaptation. By identifying the scope, ontology is

limits to contain the most relevant concepts that

regards to the application only. Scope helps to

minimize the complexity of ontology as the cost of

inference is higher for complex ontology. A

competency questions is a tool in determining the

scope. It is a basic and simple list of questions in

natural language form that need to be answered by

ontology.

4.2 Specification: Reusing Existing

Ontologies

Reusability is one of the facilities offered by

ontology to share the domain knowledge. It can

prevent an ontologist from reinventing the wheel,

hence faster development progress. Hebeler et al.,

(2009) delineate other benefits of reuse such as

simplify the domain concepts, better solution,

shorter development timeline, and focus only on

ontology modelling for a specific problem-solving

methods of the application.

In this research, relevant existing ontologies i.e.

nutritional food and cooking was considered. The

nutritional food is related to the study since our

domain is dietary menu planning, while cooking

ontology is related with a prepared food. Eight

existing ontologies that relate with the domain were

analysed from its content to determine its suitability.

Three most relevant ontologies were selected. There

are fuzzy food ontology (FFO) (Lee et al., 2010) and

type-2 fuzzy food ontology (T2FFO) (Lee et al.,

2010). FFO shows how to model the raw ingredients

Ontology Modelling of Malaysian Food Exchange List

303

of food, while T2FFO focus on the cooked food

modelling. However, these ontologies were not

available for used by others. Thus, they merely

become our references for building the ontology.

The food ontology by Cantais et al., (2005) was

imported to MyFELO as was available for others to

use.

4.3 Conceptualization: Enumerate

Important Terms

The first task in conceptualization phase is to list all

the related important terms of the domain to be

included in the ontology. The aim of this activity is

to create a comprehensive list of important terms for

the key concepts. For example, important food-

related terms are nutrients data for food – energy,

carbohydrate, protein, fat, fibre,

sodium, cholesterol and serving size;

mealtime that is suitable for food, food

allergy, cultural and religious that is

associate with a food, food groups i.e. starch,

vegetables, fruits, milk and

yogurt, meat and meat substitutes

and fats; different types of food, such as raw,

processed and cooked food, and so on.

Intermediate representations (IRs) is the tool

proposed by METHONTOLOGY (Gómez-Pérez et

al., 2004) to model the conceptualize knowledge.

The set of IRs represent the conceptual model in

tabular or graph notations. We adapt the IRs idea in

building MyFELO conceptual model.

4.4 Conceptualization: Define Class

and Class Hierarchy

This phase starts with defining the class. From the

list that obtained from the previous task, the terms

that represent a class was selected. One of the way to

identity the term as a class by looking at the term

that describe object, having independent existence

rather than terms that describe these objects (Noy

and Mcguinness, 2001).

Next, the classes were arranged into a superclass-

subclass hierarchy, or known as taxonomy. Uschold

and Gruninger (1996) proposed three approaches in

defining class hierarchy i.e. top-down, bottom-up

and middle-out. The selection of which approach to

apply is mainly depend on how an ontologist views

the domain. Since we think of foods by

differentiating the most general concept first, thus,

we opted to top-down approach. Herein class and

concept are used interchangeably.

The taxonomic relations in OD101 are referring

to the Frame ontology and the OKBC ontology

(Gómez-Pérez et al., 2004). Both of these ontologies

are knowledge representation ontologies. Among the

taxonomic relations discussed in OD101 are:

subclass-of, superclass-of, disjoint-decomposition

and instance-of. The subclass-of is based on “is-a”

relation. A class B is a subclass of A if and only if

every instance of B is also instance of A. For

example, cereal is a subclass of starch; since every

cereal grains is under starch food group. Superclass-

of relation is the inverse of subclass-of relation.

Thus, starch is the superclass for cereal. Disjoint-

decomposition relation is refers to disjoint classes,

where there is no common instances between them.

Cereal, rice and wheat was set as disjoint classes,

thus, any instance in one class cannot be an instance

of another class. OD101 also explains regarding on

the multiple inheritance, another subclass-of

taxonomic relation. This relation allows a class to be

subclass of several classes.

4.5 Conceptualization: Define

Properties of Classes

Properties are relation between two things also

known as binary relations. From the list of terms

created at step 4.3, the remaining of important terms

are probably the properties of the classes. The

examples of remaining terms are food’s energy,

carbohydrate, protein, fat, fiber,

sodium, serving size and allergy of a

food.

In OD101, for each property in the list, we need

to associate which classes it belongs. Subclass

inherits the properties of their superclass. Thus,

property should relate to the most general class that

belongs to it. However, this technique is no longer

available for Protégé. Protégé is the ontology tool

that support OD101. Thus, for this step, we refer to

the tutorial of building OWL ontologies using

Protégé 4 (Horridge, 2011) Object property and data

property are two main types of property in OWL.

The first link an individual to an individual namely

relationships between individual, while the latter

link an individual to data literal which they describe

relationships between an individual and data values.

From the remaining important terms list, we

identified the two main types of property. Object

property link between an individual with an

individual, while data property link between an

individual with data literal. For example, food’s

energy, carbohydrate, protein, fat and fiber are data

property, where the data literal is float. Allergy,

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

304

companion, served with and serving size are object

property because it links between individual food

and individual from different classes.

The naming convention for property name is

prefix with the word ‘has’ or the word ‘is’ for the

inverse property. It should start with a lower case

letter, no spaces and the second words start with

capital case letter. This naming convention helps the

property become clearer to human.

4.6 Conceptualization: Create

Instances

The final step in conceptualization phase is to create

instances of classes. Defining an instance involve

the activities of choosing a class, creating an

instance of that class and assign the object or data

properties value.

4.7 Implementation

During this phase, the conceptual model is transform

into formal model using an ontology language. The

ontology language chosen for this study is Web

Ontology Language (OWL). OWL is one of the

ontology markup languages based on eXtensible

Markup Language (XML). OWL consists of OWL

Lite, OWL DL (Description Logic) and OWL Full.

OWL DL was chosen in this research as it contains

the complete OWL vocabulary. Protégé 4.3 has been

chosen as the ontology development tool.

5 DISCUSSION

OD101 gives a good starting point to a new ontology

designer to build the ontology. It encourages a

novice to make the design decision based on their

own understanding and point of view towards the

domain. This not rigid approach gives freedom in

designing the ontology as long as the ontology

capable to fulfill the requirement of an application. It

proposes the iterative design process which allows

the beginner to start small at the ontology. Then, the

ontology is revised by assess it in the application.

The refinement is made and this evolving ontology

is continuing until it is complete. For instance, we

started the MyFELO from the class of raw and

processed food, and created the properties and

instances for the breakfast’s menu data, then the

ontology was evaluate by using the application.

After the refinement, the ontology was proceed with

the second mealtime i.e. morning snack and these

process are continue until the development is finish.

OD101 emphasizes more on the

conceptualization activity due to the two most

important tasks in ontology development contained

in it. Defining the class hierarchy and the properties

are explained in detail so that a beginner has

thorough guidelines to follow.

One good tip that can be addressed in this

method is to be consistent when using singular or

plural for naming convention of concept names.

Personally, we think singular for class name is better

because it keep simple.

The main drawback of OD101 is the missing of

management and support activities as in

METHONTOLOGY. It also lack of the IRs to model

the informal knowledge.

6 CONCLUSION

The purpose of this study was to develop the

MyFELO according to the OD101 method. This

paper has described each of the process in the

method i.e. specification, conceptualization and

implementation. OD101 proposes the iterative

approach for the ontology development. The

complexity of ontology is simplified by building an

initial version and develops it gradually along the

life cycle.

MyFELO consists of the nutritional and

personalization aspect such as race and religion. It

also includes a companion among the foods. It

covers from raw to prepared foods. This ontology is

used to make the domain assumptions explicit so

that it can leverage the burden on the processing

level of adaptation strategy in designing dietary

menu planning.

This research is still in progress. We are at the

stage of evaluation. The ontology development

already finished and the adaptation methods are

completed. To the best of our knowledge, the

reasoning process of MyFELO has successfully

adapt the menu of the most similar case to satisfy

any constraints that demand in the current problem

of dietary menu planning.

ACKNOWLEDGEMENTS

This work was conducted using the Protégé

resource, which is supported by grant GM10331601

from the National Institute of General Medical

Sciences of the United States National Institutes of

Health.

Ontology Modelling of Malaysian Food Exchange List

305

REFERENCES

Brusa, G., Caliusco, M. L. & Chiotti, O., 2007. Enabling

Knowledge Sharing within e-Government Back-Office

Through Ontological Engineering. Journal of

Theoretical and Applied Electronic Commerce

Research, 2(1), pp.33–48.

Cantais, J. et al., 2005. An example of food ontology for

diabetes control. In International Semantic Web

Conference 2005 workshop on Ontology Patterns for

the Semantic Web.

Chien-Yeh, H. et al., 2011. A Web-Based Decision

Support System for Dietary Analysis and

Recommendations. Telemedicine and e-Health, 17(2),

pp.68–75.

Gómez-Pérez, A., Fernández-López, M. & Corcho, O.,

2004. Ontological Engineering : with examples from

the areas of Knowledge Management, e-Commerce

and the Semantic Web. Second. X. Wu & L. Jain, eds.,

Springer-Verlag London.

Hebeler, J. et al., 2009. Semantic Web Programming M.

Dean, ed., Indiapolis, Indiana: Wiley Publishing, Inc.

Horridge, M., 2011. A Practical Guide to Building OWL

Ontologies using Protege 4 and CO-ODE Tools,

p.107.

Kim, E., 2012. Korean Food Ontology. In Seventh

International Conference on Formal Ontology in

Information Systems. p. Poster Presentation.

Kovacic, K. J., 1995. Using Common-Sense Knowledge

for Computer Menu Planning. Case Western Reserve

University.

Lee, C., Wang, M., Acampora, G., et al., 2010. Diet

Assessment Based on Type-2 Fuzzy Ontology and

Fuzzy Markup Language. , 25, pp.1187–1216.

Lee, C., Wang, M. & Hagras, H., 2010. A Type-2 Fuzzy

Ontology and Its Application to Personal Diabetic-

Diet Recommendation. IEEE Transactions on Fuzzy

Systems, 18(2), pp.374–395.

Marling, C. R., 1996. Integrating Case-Based and Rule-

Based Reasoning in Knowledge-Based Systems

Development. Case Western Reserve University.

Marling, C. R., Petot, G. J. & Sterling, L. S., 1999.

Integrating Case-Based and Rule-Based Reasoning to

Meet Multiple Design Constraints. Computational

Intelligence, 15(3), pp.308–332.

Noah, S. A. et al., 2004. DietPal : A Web-Based Dietary

Menu-Generating and Management System. Journal

of Medical Internet Research, 6(1), p.13.

Noy, N. F. & Mcguinness, D. L., 2001. Ontology

Development 101 : A Guide to Creating Your First

Ontology. In Stanford Medical Informatics Technical

Report SMI-2001-0880. pp. 1–25.

Pizzuti, T. et al., 2014. Food Track & Trace ontology for

helping the food traceability control. Journal of Food

Engineering, 120, pp.17–30.

Saludin, F. A., Zakaria, N. & Husain, W., 2010. User

Requirement Analysis for Personalized Cancer Dietary

Planning and Menu Construction. In IEEE Conference

on Biomedical Engineering & Sciences

. pp. 410–416.

Tee, E. S. et al., 1997. Nutrient Composition of Malaysian

Foods 4th Editio., Kuala Lumpur: Institute for

Medical Research.

Tumnark, P., Oliveira, L. & Santibutr, N., 2013.

Ontology-Based Personalized Dietary

Recommendation for Weightlifting. In 2013

International Workshop on Computer Science in

Sports. Atlantis Press, pp. 44–49.

Uschold, M. & Gruninger, M., 1996. Ontologies :

Principles , Methods and Applications. Knowledge

Engineering Review, 11(February), p.50.

Warshaw, H. S., 2010. Diabetes Meal Planning Made

Easy 4th ed. R. Renshaw, ed., American Diabetes

Association, Inc.

KEOD 2015 - 7th International Conference on Knowledge Engineering and Ontology Development

306