Discovering Ontological Knowledge in Unstructured Recipes of a

Portuguese Monk from the 16th Century

Orlando Belo

, Bruno Silva

and Anabela Barros

ALGORITMI Research Centre/LASI, University of Minho, 4710-059 Braga, Portugal

CEHUM, Centre for Humanistic Studies, University of Minho, 4710-059 Braga, Portugal

Keywords: Ontology Learning, Knowledge Extraction, Unstructured Textual Data, Natural Language Processing,

Linguistic Ontologies, Graph Databases.

Abstract: Ontology learning is often applied to textual data sources with the aim of identifying, extracting and

representing their various data elements, as well as their semantic relationships. Ontologies are excellent

instruments for the representation of knowledge about one or more domains of knowledge, which enable us

to study in detail the knowledge of the domain they host. With this in mind, we devised and developed a semi-

automatic ontology learning system. It was specifically oriented for discovering the knowledge contained in

a set of ancient texts of culinary recipes of a monk of the 16th century. Using the system, we produced an

ontology incorporating a large diversity of culinary elements and their relationships, which offer a very rich

field of research of the culinary of the 16th century in Portugal – the ontology was exposed and explored using

the native mechanisms of a graph database management system.

1 INTRODUCTION

Over the past few years, ontologies have probably

been one of the most widely used models of

knowledge representation in real-world applications,

particularly in organizations that need to acquire and

manage knowledge about one or more application

domains (van Eijnatten, 2004). Since the emergence

of the Semantic Web (Ding et al., 2007), ontological

models have been catapulted into the spotlight,

coming to be widely used in various computational

systems (Davies et al., 2003). Since then, several

studies have been carried out in the area of ontologies,

due to the determining role they play in data

organization and in the representation of the

relationships established with it, in any area or

domain of application.

An ontology (Guarino et al., 2009) (Keet, 2018)

defines a set of concepts related to each other, with

the purpose of creating interoperability between them

and forming a semantic link to the domain in

question. The semantics illustrated by an ontology

consist of concepts, relationships, and properties.

https://orcid.org/0000-0003-2157-8891

https://orcid.org/0009-0000-5760-3193

https://orcid.org/0000-0002-2959-9200

Concepts relate to each other through the

establishment of semantic relationships that

characterize the way they can be conjugated. In this

way, it is possible to work and reason about the

context and application field of concepts that an

ontology welcomes, creating a real perception of

what the domain presents. Ontologies have a great

application level, being transversal to any existing

problem in the context of the real world. Application

areas can range from health sciences to e-commerce,

or any other area that needs to explore shareable and

reusable information within a particular knowledge

domain. In addition to being a scalable and

sustainable way of storing information, an ontology

aims at better management of the knowledge

extracted from a knowledge domain. An ontology

allows for faster analysis and under-standing of what

is exposed in the source of knowledge. The process

of conceiving an ontology often referred as ontology

learning (Cimiano et al., 2009) (Asim et al., 2018),

encompasses all the tasks we use to perform to extract

the terms, concepts, relationships, rules or axioms

from textual sources to get the structure of an

200

Belo, O., Silva, B. and Barros, A.

Discovering Ontological Knowledge in Unstructured Recipes of a Portuguese Monk from the 16th Century.

DOI: 10.5220/0012189500003598

In Proceedings of the 15th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2023) - Volume 2: KEOD, pages 200-207

ISBN: 978-989-758-671-2; ISSN: 2184-3228

ontology. Among all the proposed methods for

ontology learning on texts, the proposal of Brewster

(2006) stands out significantly, given the simplicity

of its task structure and application model. However,

the nature of an ontology learning process is quite

complex, usually requiring the application of very

sophisticated processing techniques, which involve

frequently natural language processing techniques

(Sharma, 2021) and machine learning algorithms

(Zhang et al., 2018).

In this paper, we present and discuss a semi-

automatic ontology learning system. It was conceived

specifically for extracting a culinary ontology from a

textual source of the 16th century, which were

collected by the monk José Joaquim de Santa Teresa,

recorded in the library of the Monastery of S.

Martinho de Tibães, Braga, Portugal, and published

by Barros in "As Receitas de um Frade Portuguese no

séc. XVI" (“The Recipes of a Portuguese Monk in the

16th century”) (Barros, 2013). The referred data

source hosts about three hundred recipes, in disparate

formats, their ingredients and preparation processes,

which are very important elements for the study of

Portuguese cuisine during this time. The ontology

produced incorporates a large diversity of culinary

elements and their relationships, offering a very rich

field of research for students, professors, researchers

and individual users and providing a very interesting

view of the culinary practices of 16th in Portugal.

Additionally, we will present the process of

discovering and extraction the ontology we carried

out, as well as describe all its main tasks and

techniques used, from the preparation of the texts of

the recipes to their processing and generation of the

ontology. Next, section 2 presents and discusses

several aspects related to ontology learning aspects,

section 3 introduces the book of recipes, presenting

and describing our application case, and exposes how

we extracted the ontology of recipes, and, finally,

section 4, presents some conclusions and future work.

2 RELATED WORK

Ontologies (Guarino et al., 2009) (Keet, 2018) have

long proven their usefulness and versatility in

knowledge acquisition and management. Currently,

its application is very widespread. We can find

ontological systems in the most varied fields of

application, such as medicine, management,

economics, or linguistics. Gruber (1995) proposed the

most consensual definition in the field of Computer

Science for an ontology, saying that it “is an explicit

specification of a conceptualization”. However, many

other authors proposed their own definitions of

ontologies, such as presented in (Borst, 1997) or

(Uschold and Gruninger, 1996). However, all of them

based their definition in a specification of a shared

conceptualization of the knowledge of one or more

specific application domains. They are very important

instruments in processes of analysis and application

of knowledge. However, contrary to the process of

exploring the knowledge of an ontology, all those

who are or have been involved in a process of

defining and characterizing an ontology know that

this is not an easy process. In order for us to be able

to successfully learn an ontology, we have to have at

our disposal credible data sources, as well as people

with the expertise and knowledge in the domain of the

knowledge involved and in the process of building the

required ontology, whether manual or automatic.

The manual construction of an ontology is often

too time-consuming for experts in the field, as it

requires a broad analysis of the context concerned and

the corresponding identification of all the elements

relevant to the ontological constitution. The research

of this theme, nowadays, falls mainly on the semi-

automatic and automatic construction, in order to

overcome the problems identified in the process of

manual construction of the ontology. The processes

of ontology learning (Mishra and Jain, 2014) aim to

represent the semantic relationships between the

various elements present in a data set. Today, it is

common to choose to carry out this work in textual,

unstructured data sources, given the enormous

richness of its content (Choudhary and Tomar, 2014)

(Belhoucine and Mourchid, 2020). In this domain,

one of the great purposes of ontologies is to identify

and characterize relevant elements in the contents of

texts, as well as to establish the relationships between

them. Then, through control and automation

mechanisms, ensure the execution of fast and

effective analysis processes of the contents addressed

and treated in these texts.

In (Brewster, 2006) it was used an image of a

“layer cake” for representing the order that an

ontology learning process must execute the extraction

tasks (Figure 1). The author ordered the extraction

tasks based on their difficulty to perform. The

Brewster’s proposal puts on the base of the cake the

most simply task, the identification of terms (task 1)

containing in the texts. In the second and third layer

appears, respectively, the tasks of synonyms (task 2)

and concept identification (task 3), which in some

cases can be performed simultaneously. Next, we

need to establish the concept hierarchies (task 4),

followed by the identification of relationships (task 5)

between concepts. Finally, at the top level, appears

Discovering Ontological Knowledge in Unstructured Recipes of a Portuguese Monk from the 16th Century

201

Figure 1: The tasks of an ontology learning process according to (Brewster, 2006).

the task for discovering rules (task 6). As happens in

other fields, this is not an exclusive method for

organizing an ontology learning process. To note,

(Tiwari and Jain, 2014) added some value to the

proposal done in (Brewster, 2006) saying that some

of the tasks can be executed simultaneously, which

speed up the extraction process time. Later, in (Asim

et al., 2018) were added some new tasks to the

proposal made in (Tiwari and Jain, 2014). It added to

the process two very important tasks: data pre-

processing and ontology evaluation. They formally

included two tasks we really do in practice before

starting and after executing a conventional extraction

process.

In the field of cooking, several proposals for

ontologies have been presented, some of them quite

specific. For example, we have the works of (Noy and

McGuinness, 2001) or (Graça et al., 2005), which

presented ontologies for the enology field, or

(Markantonatou et al., 2021), which modelled the

domains of dishes that figured in a large number of

menus quite diverse. In the field of cooking recipes,

we find also several interesting examples of

ontologies. For example, in (Villarías, 2004), it is

presented an ontology for cooking recipes to be used

by a Web semantic querying system, or in (Batista et

al., 2006) it was devised and implemented an

ontology for the cooking domain, involving concepts

such as actions, food, recipes, and utensils. As well,

in (Monica et al., 2014) is presented an ontology for

culinary processes defined in recipes and the authors

of (Nanba et al., 2014) created an ontology for the

establishment of hyponymy, synonymy, attributes,

and meronymy in cuisine recipes. The definition of

another ontology for cooking recipes was not the

motivation of this work, but rather the extraction

process implied from the nature of a set of texts

written in classical Portuguese of the 16th century.

We wanted to demonstrate the application of various

techniques, models and tools used today in ontology

learning, and apply them over ancient textual data

sources. In particular, on a set of very peculiar texts:

cuisine recipes. Then, we want to verify whether the

culinary processes and the ingredients used at the

time would be similar (or not) to those we use today.

In addition, the interest in carrying out this work was

reinforced by the utility that an ontology of the 16th

century cuisine has for all those who develop their

research work in the field of Food History in Portugal.

In the next section, we will present and discuss the

process of discovering ontological knowledge in

unstructured recipes of a Portuguese monk from the

16th century.

3 DISCOVERING THE RECIPE

ONTOLOGY

3.1 The Monk’s Recipes

The analysis and editing of a vast set of ancient

cooking recipes, from the second oldest culinary

manuscript known in Portugal (the first notebook of

codex 142 of the District Archive of Braga),

originated a source of textual data, containing very

rich and diverse information about Portuguese cuisine

in the 16th century. The manuscript contains the texts

of the recipes, referring ingredients and a diverse set

of combinations of culinary processes. A monk, José

Joaquim de Santa Teresa, collected the recipes. He

decided to gather various manuscripts scattered

throughout its congregation, and proceeded to register

them in the library of the Monastery of S. Martinho

de Tibães (Braga, Portugal). In 2013, Anabela Barros,

edited and published the recipes in the book “As

Receitas de um Frade Português no séc. XVI” (“The

Recipes of a Monk Portuguese in the century. XVI")

(Barros, 2013). This book provides a rich source for

studying the various culinary elements – recipes,

ingredients, methods, processes, etc. –, included and

dispersed throughout the various unstructured texts of

the source. All these elements are very important for

studying the 16th century Portuguese cuisine. As

already referred, the creation and development of a

domain ontology, specifically oriented to the

reception of the knowledge associated with these

elements would provide a very useful instrument for

KEOD 2023 - 15th International Conference on Knowledge Engineering and Ontology Development

202

all who wanted to learn about the cuisine that was

taking place in that century. In a first stage of our

work, we studied the texts of the recipes, both in terms

of contextual and structural level – Table 1 presents

an example of a recipe – “Sopas de Vaca

Contrafeitas” (“Counterfeit Cow Soup”). The aim

was establishing a knowledge base and understanding

writing patterns, as well as characterizing the natural

language used in the texts.

Table 1: The text of a recipe in classic Portuguese.

Recipe Nr.

and Title

(Block 1)

Ingredients

(Block 2)

Method

(Block 3)

Next, we defined the process of extraction of the

ontology, the way we would carry it out, taking into

account the particularities that the texts presented,

such as writing patterns, linguistic structure,

frequency of terms, and heterogeneity of the texts.

Finally, we exported the ontological structures

created to a graph database management system.

3.2 Extracting the Ontology

The process of extracting an ontology from a set of

unstructured texts is not easy to accomplish (Tiwari

and Jain, 2014). Despite the numerous computational

tools that exist today, its automation is not yet

completely automatic. Each process of development

of an ontology has its own specificities, as well as the

texts that serve as its source of knowledge. Usually,

in practice these processes use to involve a hybrid

approach, involving work performed by experts in the

domain of ontology (the manual part) and work

performed by natural language processing tools, text

mining and machine learning (the automatic part).

Although the automatic part is very attractive and

saves temporal and human resources in cases of large

textual data sources, the manual approach is always

recommended for ensuring the quality of the

ontology, which depends a lot on the complexity and

granularity of the data elements included in the texts

(Wong et al., 2012).

In the development of the ontology learning

process, we followed a supervised approach,

particularly in the initial part of the ontology

definition, leaving the automatic part of the process to

perform syntactic analysis and term extraction tasks.

In any case, we follow the approach proposed in

(Brewster, 2006), organizing our process in such a

way as to perform the tasks of extracting terms,

synonyms, concepts, hierarchies, relationships and

rules in the way he indicated. Thus, we first extracted

all the terms present in the texts of the recipes, with

the aim of creating a set of terms to be cataloged. We

then grouped each of the identified terms with all their

synonyms found in the previous set. This

significantly increased the accuracy of the concept

identification process and greatly reduced the number

of redundant terms. After having carried out the task

of grouping terms, we associated them with concepts

for establishing a concrete correspondence between

the language of the domain of knowledge and the

ontological system to be created. Then, we

established the hierarchical relationships and defined

the relationships between concepts, specifying the

domain and the scope of the relationship. However,

in order to carry out successfully all the extraction

tasks, we had to analyze preliminarily, in detail, the

texts of the recipes, in order to organize the extraction

process and facilitate the treatment of the data.

Therefore, we decided to divide each recipe into three

distinct parts (Table 1): block 1, which contains the

identifier and designation of the recipe; block 2,

which has a description of the ingredients used by the

recipe and, sporadically, some additional notes; and

block 3, which describes the recipe preparation

process. It should be noted that the information in

block 2 does not appear explicitly in the manuscript,

having been prepared previously by the author of

(Barros, 2013) during the edition of the manuscript,

who presented them in modern Portuguese.

Table 2: Some concepts identified in the recipes.

Concept Description

1 Recipe Represents each of the culinary records or texts for

the confection of a dish and its general

characteristics.

2 Ingredient Identifies an ingredient and some additional notes.

3 Note Represents a culinary note, made by the editor (who

removed it or deduced it from the text of the recipe)

and inserted in the block o

redients.

4 Procedure Describes all the steps (processes) of the

aration o

the reci

5 Process Represents each process identified in the recipe

preparation block.

6 Index Allows for the enumerating all the words present in

the texts o

the reci

es.

The existence of this block greatly facilitated the

process of identifying ingredients in the recipes.

Originally, these elements figured in very variable

positions throughout the manuscript. The definition

of the most pertinent classes in culinary ontology

depends highly on the representative units that a

recipe presents. Initially, we recognized and

established six classes (concepts) to incorporate the

Discovering Ontological Knowledge in Unstructured Recipes of a Portuguese Monk from the 16th Century

203

entities referred in each text. Table 2 presents and

describes the concepts we identified – nothing new if

we analyze some of the current cuisine recipes

ontologies. The formalization of an ontology is often

a too complex process to do through logical

definitions formally written. To overcome the need to

simplify the entire process of creation and evolution

of an ontology, several tools were developed having

the ability for expressing this formalization in an

organized. Next, we will see how the remaining tasks

of the ontology extraction process developed,

identifying whenever necessary the tools that were

being used.

Textual documents have their own characteristics

and specificities, which vary according to temporal,

external and personal factors. Just as a human being

has fingerprints, a person who writes a text somehow

incorporates his own identity into the writing of the

text. It is something like a signature. To recognize this

kind of signature, we need to study how the sentences

in the text are articulated and constructed

syntactically. To extract the syntactic and lexical

elements present in the writing of a given author, we

can define a set of specific patterns to identify

linguistic marks (terms), that have been used by the

author of the texts of the recipes. Subsequently, these

terms have to be evaluated to determine whether they

are acceptable or of interest in the context of the

ontology of cooking recipes. To establish these

patterns we relied on the study of the various recipes

we had available.

Table 3: Examples of patterns used for identifying terms.

Pattern Sentence Process

1 [^\s]+-[^\s\.]* “Deitar-se-á em um

tachinho...”

deitar

2 [^\s,\.]+ [^\s,\.]+ “...salsa e cebola

picados,...”

picar

3 [^\s,\.]+ “...e clarificado com

uma clara de ovo...”

clarificar

4 [^\s,\.]+

[^\s,\.]+|se

[^\s,\.]+

[^\s,\.]+|lhe

s,\.]+ [^

.]+

“...a qual se lhe lançará

muito bem...”

lançar

5 [^\s,\.]+ com

s,\.]+

“...e nela se frigirão com

azeite...”

frigir

6 [^\s,\.]+ em

s,\.]+

“...e afogados em azeite

se lançará...”

afogar

7 [^\s,\.]+| lhe

s,\.]+

“...um fogo brando se

derreterá nela...”

derreter

8 [^\s,\.]+ [^\s,\.]+ “...havendo primeiro

sido passadas...”

passar

In this context, we identified several patterns,

some more generic than others. We defined patterns

using specific regular expressions. For example, the

pattern '[^\s]+-[^\s\.] *' allows for identifying all

verbs with associated pronouns in a sentence. During

the process of syntactic analysis of the procedure of a

recipe, we found that, generally, a verb followed by a

post verbal pronoun is a culinary process. These cases

arise essentially at the beginning of sentences.

However, this particularity was not considered in the

expression, so that it is in some way “permissive”.

The sentence "Deitar-se-á em um tachinho..." ("It will

lie down in a pot...") was one of the phrases that may

be instantiated with the referred pattern, in which we

can easily identify the culinary process “deitar” (to

put). Another pattern we defined was '[^\s,\.] + [^\s,\.]

+', which allows the extraction of two words, one

relating to ingredients and the other relating to the

state of the ingredients. This pattern defines a state

that is interpreted as a verb in participle form.

Through this pattern, the phrase "...salsa e cebola

picados..." ("...chopped parsley and onion...")

allowed to identify the “picar” (to chop) process.

Many other processes, such as “clarificar” (to clarify),

“lançar” (to cast), or “frigir” (to fry), have been

identified by other patterns in sentences such as “...e

clarificado com uma clara de ovo...” (“... and clarified

with an egg white...”), “...a qual se lhe lançará muito

bem...” (“... which will be thrown at you very well...”,

or “...e nela se frigirão com azeite...” (“...and in it they

will be fried with olive oil...”), respectively. Other

examples of patterns can be found in Table 3.

After extracting the terms with the patterns, it was

necessary to classify syntactically the content of the

preparation (method) of a recipe. We made the

tokenization of the text of the culinary procedure

using Spacy (Honnibal and Montani, 2017). Once all

the sentences were classified, it was possible to

extract the syntactic function assigned to each word,

the tokens. This assignment is critical to shaping the

rules for acceptance of the terms, in conjunction with

the established standards. Each word was instantiated

in a given object, whose properties can be its literal

meaning, the reduction of the word to its root or its

grammatical class, that is, the attribute that allows for

identifying the verbs. After segmenting the phrases of

the texts in tokens, we associated with each identified

process the various ways in which it appeared written

in the recipes. In practice, we mapped all the variants

of a culinary process, taking into account their verbal

form, gender and number. In this way, it was easier to

identify the occurrences of a process in a given recipe.

Next, we extracted concepts and aggregated

synonyms. The attribution of concepts (or classes)

(Table 2) to the extracted terms was, to some extent,

quite evident, since their location in the text of the

recipe revealed their identity in most cases. The

elements contained in block 1 of the recipe (Table 1)

KEOD 2023 - 15th International Conference on Knowledge Engineering and Ontology Development

204

are integrated into the concept “Receita” (“Recipe”).

This block always contains the same elements:

number and name of the recipe. However, the same is

no longer the case in the other blocks. At least so

directly, since the extraction of the concepts

associated here depended directly on the algorithms

used for extracting terms, relationships and associated

rules.

At this point, we start the automatic part of the

ontology learning process. As we process the recipe,

we created its identity. Using the contents of block 2

of a recipe (Table 1), we can find out which entities

belong to the concept “Ingredient”. In some cases, in

block 2, may appear references to entities that belong

to the concept “Note”, since they are do not belong to

the recipe or to the list of ingredients. However, they

indirectly help the preparation of the recipe. Then, in

addition to the ingredients, extracted manually by the

text editor and placed in block 2, using contents of

block 3, we can infer the identification the most

representative elements included in the “Method”

concept description, although may arise other

processes derived from the processing of the content

of this class. Finally, we identified the representative

entities of the “Process” concept.

If no process was identified, we assumed that we

are in the presence of a poor or incomplete procedural

description of a recipe, an incomplete method. Any

process identified by the system is an abstraction of a

more specific definition, such as an action performed

by a person preparing a recipe or, simply, by a

procedure merely exclusive to the domain of cooking.

For example, it is correct to say that the “deitar”

(“throwing”) process allows for inferring an action of

the user and that the “ferver” (“boiling") process

represents an action relative to an ingredient.

However, this differentiation was not explicitly

integrated into the ontology, since the defined

patterns were not developed to make this hierarchical

distinction.

To finish the process of extracting the ontology,

we needed to confirm the identity of each term, as

well as the role it will play in the semantic area.

Relationships allow for establishing chains of

knowledge and affinity between two entities, linking

them through one or more characteristics. Like the

properties of a concept, the relationship is an integral

part of the process of determining the role that a given

element plays in the ontology. When it is not possible

to ascertain the properties of a particular element, it is

possible to make the recognition of the entity

concerned if we identified a relationship. In the

recipes we analyzed, all the relationships between

concepts were created by the terms found, and all the

terms found relate to the concept “Receita”

(“Recipe”). This logic is justified by the fact that all

the elements found are a part of a recipe and by the

direct relationship between a given concept and all

the recipes introduced into the system. Each concept

found is about a recipe, in which the relationship

assumes a degree of possession and integral part.

When the domain is "Recipe", the relation is

expressed as a possessive bond, and when this is

characterized as the initial set, the relationship is

identified as a part of the recipe. Figure 2 shows a

conceptual definition of the ontology we extracted.

The ontology’s structure is simple and very similar to

the other recipe ontologies we studied before.

Concepts ::

RECIPE (Id,

Description);

INGREDIENT

(Description);

NOTE (Description,

Contents);

PROCEDURE

(Description);

PROCESS (Name,

Variations);

INDEX (Name);

Relationships ::

RECIPE has INGREDIENT;

INGREDIENT of_a RECIPE;

RECIPE has NOTE;

NOTE of_a RECIPE;

RECIPE has PROCEDURE;

PROCEDURE of_a RECIPE;

RECIPE has PROCESS;

PROCESS of_a RECIPE;

PROCEDURE has PROCESS;

PROCESS of_a PROCEDURE;

RECIPE has INDEX;

INDEX of_a RECIPE;

Figure 2: A conceptual definition of the recipe ontology.

3.3 Exploring the Ontology

After we finished the extraction process, we tried to

preserve the ontological structures created. The

preservation of these structures was carried out in a

graph database management system, the Neo4J

(Neo4J, 2023). Through it, we can perform all the

preservation operations we need, as well as explore

the stored ontology through its interrogation

language: Cypher. The registration of the extracted

entities can be done through a combination of pre-

existing data with the knowledge to be imported. If

there are no references yet, all the new information in

the ontology is inserted.

The improvement of the ontology was made after

the import of the ontological structures into the Neo4J

system, and was carried out in two different ways:

through the database system or through an OWL file.

The difference between these two ways of improving

the ontology is only in terms of semantic inference.

In Neo4j it is done using queries and in OWL using

HermiT (Glimm et al., 2014). Hermit allows

completing any information that may not have been

associated with a given concept. However, it is

extremely time-consuming, especially if the

determination of the new relationships is made after

the insertion of several records.

Discovering Ontological Knowledge in Unstructured Recipes of a Portuguese Monk from the 16th Century

205

To allow the dissemination of results and the

experimentation of the ontological system created, we

designed and developed a specific Web platform. The

development of this platform was done through the

Flask micro framework, written in Python and

architected based on an MVC (Model-View-

Controller) pattern, not only for the organization of

the application between the logical layer and the

graphic layer that this pattern establishes, but also for

the ease it gives us to add or modify system

functionalities. The Web platform developed allows

for exposing in a simplified way all the information

of the ontology extracted. In a real application

scenario, this platform substantially facilitates the

process of study and research, both of a recipe and its

components, as well as of a component and all the

recipes linked to it. In Figure 3, we can see a small

view of the ontology in Neo4J environment.

Figure 3: A view of the ontology provided by the Neo4J

system.

4 CONCLUSIONS AND FUTURE

WORK

In this paper, we presented a semi-automatic ontology

learning system conceived for extracting a culinary

ontology from a large set of recipes of the 16th

century published in (Barros, 2013). The

implementation of the system covered all the stages

of an ontology extraction process, from text

preprocessing to the materialization of the structures

of ontology. In addition, it provides a Web platform

for consulting and analyzing the knowledge

incorporated in the ontological system. This platform

is a very useful tool for all those who wish to study

the Portuguese cuisine of the 16th century, through

the knowledge extracted from the cooking recipes

collected. It facilitates a lot the process of research

and analysis, whether of a recipe and its components,

or of a component and all the recipes related to it. In

semantic terms, the ontology allows for establishing

a bridge between the Portuguese culinary processes

of the 16th century and the processes of the modern

era that, curiously, in one way or another, are similar

or, sometimes, the same. An identical reasoning can

be applied to the ingredients that were used then in

recipes, but that today are already replaced by others.

This ontology, validated by the author of (Barros,

2013), also provides means to identify culinary

patterns of the 16th century and compare them with

current culinary practices, which we believe that is

valuable for researchers in the area of Food History,

which are dedicated to investigating and comparing

them over time.

The version of the ontology we have today is not

complete. However, it reflects adequately the

knowledge contained in the recipes gathered by the

Portuguese monk José Joaquim de Santa Teresa, in

the 16th century. One of the improvements we plan to

make in a near future involves the development of

extraction mechanisms that will allow us to

distinguish what are cooking processes and user

actions. This is because we found that some of the

verbs used in the recipes identify processes inherent

in the ingredients and others correspond to steps that

must be followed by the person who is making the

dish. This distinction will impose a more precise and

detailed syntactic analysis, since we will need to

analyze all the elements that precede or succeed a

verb.

ACKNOWLEDGEMENTS

This work has been supported by FCT – Fundação

para a Ciência e Tecnologia within the R&D Units

Project Scope: UIDB/00319/2020, and the PhD grant:

2022.12728.BD.

REFERENCES

Asim, M., Wasim, M., Khan, M., Mahmood, W., Abbasi,

H. (2018). A survey of ontology learning techniques

and applications. Database, Volume 2018. DOI:

https://doi.org/10.1093/database/bay101.

Barros, A. L. (with Seiça, M., Veloso, J. & Aguiar, M.)

(2013). As receitas de cozinha de um frade português

KEOD 2023 - 15th International Conference on Knowledge Engineering and Ontology Development

206

do século XVI. Coimbra: Imprensa da Universidade de

Coimbra.

Batista, F, Pardal, J., Mamede, N., Vaz, P., Ribeiro, R.

(2006). Ontology construction: cooking domain.

Technical Report, INESC-ID, February.

Belhoucine, K., Mourchid, M. (2020). A Survey on

Methods of Ontology Learning from Text. pages 113–

123.

Borst, W. (1997). Construction of engineering ontologies

for knowledge sharing and reuse. Twente, sep.

Brewster, C. (2006). Ontology Learning from Text:

Methods, Evaluation and Applications Paul Buitelaar,

Philipp Cimiano, and Bernado Magnini (editors) (DFKI

Saarbrücken, University of Karlsruhe, and ITC-irst),

Amsterdam: IOS Press (Frontiers in Artificial

Intelligence and appl. Computational Linguistics,

32(4):569–572. DOI: https://doi.org/

10.1162/coli.2006.32.4.569

Choudhary, J., Tomar, D. (2014). Semi-Automated

Ontology building through Natural Lan-guage

Processing. International Journal of Computers &

Technology, 13(8), 4738–4746. DOI:

https://doi.org/10.24297/ijct.v13i8.7072.

Cimiano, P., Mädche, A., Staab, S., Völker, J. (2009).

Ontology Learning. In: Staab, S., Studer, R. (eds)

Handbook on Ontologies. International Handbooks on

Information Systems. Springer, Berlin, Heidelberg.

DOI: https://doi.org/10.1007/978-3-540-92673-3_11.

Davies, J., Fensel, D., van Harmelen, F. (2003).

Conclusions: Ontology-driven Knowledge

Management – Towards the semantic web?, Towards

the Semantic Web, pp. 265–266, John Wiley & Sons,

Ltd.

Ding, L., Kolari, P., Ding, Z., Avancha, S. (2007). Using

Ontologies in the Semantic Web: A Survey. In:

Sharman, R., Kishore, R., Ramesh, R. (eds) Ontologies.

Integrated Series in Information Systems, vol 14.

Springer, Boston, MA. https://doi.org/10.1007/978-0-

387-37022-4_4

Glimm, B., Horrocks, I., Motik, B., Stoilos, G., Wang, Z.

(2014). HermiT: An OWL 2 reasoner. Journal of

Automated Reasoning, 53(3), 245–269.

Graça, J., Mourao, A., Duarte, M., Anunciação, O.,

Monteiro, P., Pinto, S., Loureiro, V. (2005). Ontology

building process: The wine domain. 5th Conference of

the European Fed-eration for Information Technology

in Agriculture, Food and Environment.

Gruber, T. (1995). Toward principles for the design of

ontologies used for knowledge sharing? International

Journal of Human-Computer Studies, 43(5-6):907–928,

nov. DOI: https://doi.org/10.1006/ijhc.1995.1081.

Guarino, N., Oberle, D., Staab, S. (2009). What Is an

Ontology?, In Staab, S., Studer, R. (eds) Handbook on

Ontologies. International Handbooks on Information

Systems. Springer, Berlin, Heidelberg. DOI:

https://doi.org/10.1007/978-3-540-92673-3_0 2009

Honnibal, M., Montani, I. (2017). spaCy · Industrial-

strength Natural Language Processing in Python.

Homepage, https://spacy.io/api, last accessed

2023/06/22.

Keet, M. (2018). An Introduction to Ontology Engineering,

University of Cape Town. Homepage,

https://people.cs.uct.ac.za/~mkeet/OEbook/, last

accessed 2023/05/05.

Markantonatou, S., Toraki, K., Minos, P., Vacalopoulou,

A., Stamou, V., Pavlidis, G. (2021). "AMAΛΘΕΙA: A

Dish-Driven Ontology in the Food Domain" Data 6, no.

4: 41. DOI: https://doi.org/10.3390/data6040041

Mishra, T., Jain, S. (2014). Automatic Ontology

Acquisition and Learning. International Jour-nal of

Research in Engineering and Technology. 03. 38-43.

Monica, S., Krisnadhi, A., Cong, W., Gallagher, J., Hitzler,

P. (2014). An Ontology Design Pat-tern for Cooking

Recipes. In Proceedings of the 5th Workshop on

Ontology and Se-mantic Web Patterns (WOP2014) co-

located with the 13th International Semantic Web

Conference (ISWC 2014), Riva del Garda, Italy,

October 19.

Nanba, H., Doi, Y., Tsujita, M., Takezawa, T., Sumiya, K.

(2014). Construction of a cooking ontology from

cooking recipes and patents. In Proceedings of the 2014

ACM Interna-tional Joint Conference on Pervasive and

Ubiquitous Computing: Adjunct Publication (UbiComp

'14 Adjunct). Association for Computing Machinery,

New York, NY, USA, 507–516. DOI:

https://doi.org/10.1145/2638728.2641328.

Neo4J Graph Data Plataform, Homepage (2023).

https://neo4j.com/, last accessed 2023/06/22.

Noy, N., Mcguinness, D. (2001). Ontology Development

101: A Guide to Creating Your First Ontology.

Knowledge Systems Laboratory. 32.

Sharma, A. (2021). Natural Language Processing And

Sentiment Analysis, In International Re-search Journal

of Computer Science, 8(10):237. DOI: https:/doi.org/

10.26562/irjcs.2021.v0810.001

Tiwari, S., Jain, S. (2014). Automatic Ontology Acquisition

and Learning. International Journal of Research in

Engineering and Technology, 03(26):38–43, Nov. DOI:

https://doi.org/10.15623/ijret.2014.0326008

Uschold, M., Gruninger, M. (1996). Ontologies: principles,

methods and applications. The Knowledge Engineering

Review, 11(2):93–136, jun. DOI:

https://doi.org/10.1017/S0269888900007797.

van Eijnatten, F., Putnik, G. (2004). Chaos, Complexity,

Learning, and the Learning Organiza-tion, The

Learning Organization, 11(6), pp. 418-429.

Villarías, L. (0420). Ontology-based semantic querying of

the web with respect to food reci-pes. Master’s thesis,

Technical University of Denmark.

Wong, W., Liu, W., Bennamoun, M. (2012). Ontology

learning from text. ACM Computing Surveys, 44(4):1–

36, Aug. DOI: https://10.1145/2333112.2333115.

Zhang, L. (2018). Wang, S., Liu, B., Deep Learning for

Sentiment Analysis: A Survey. WIREs Data Mining

Knowledge Discovery, Volume 8, Issue 4, July/August.

DOI: https://doi.org/10.1002/widm.1253

Discovering Ontological Knowledge in Unstructured Recipes of a Portuguese Monk from the 16th Century

207