Evaluating and Exploring Text Fields Information Extraction into

CIDOC-CRM

Davide Varagnolo

, Guilherme Antas

, Mariana Ramos

, Sara Amaral

, Dora Melo

2,3 a

and Irene Pimenta Rodrigues

1,3 b

Department of Informatics, University of

Evora, Portugal

Polytechnic of Coimbra, Coimbra Business School—ISCAC, Coimbra, Portugal

NOVA Laboratory for Computer Science and Informatics, NOVA LINCS, Portugal

Keywords:

Natural Language Processing, Knowledge Representation, Knowledge Discovery, Semantic Web, Archives

Linked Data Semantic Representation.

Abstract:

This paper presents a method for extracting information from ISAD(G) elements, that contain semi-structured

text descriptions. Natural language processing is done using Gate environment and deﬁning the set of Jape

rules necessary to process the text and extract the intended information. The evaluation of the information

extraction processes is done in a sample of 800 records for each type of information, and a dataset that is

manually built for each type of information considered, such as baptisms, passport requisitions testaments,

etc. The implementation of several automatic information extraction processes enables the population of

the CIDOC-CRM knowledge base with new linked events and entities automatically. The exploration of

the information, migrated from DigitArq and extracted from text descriptions represented in CIDOC-CRM,

is done through SPARQL queries enabling new visualisations of the archival records and the retrieval of

information collected in different records from different archives.

1 INTRODUCTION

EPISA (Entity and Property Inference for Seman-

tic Archives) is a research project involving the Por-

tuguese National Archives - Torre do Tombo, archival

experts, and Information and Computer Science re-

searchers The project aims to design a prototype, as

an open-source knowledge platform, aiming to rep-

resent archival information on a linked data model.

One of the project’s major tasks is the semantic mi-

gration, i.e, the process to extract and represent the

relevant entities and their properties from the exist-

ing records in the actual DigitArq, (Ramalho and Fer-

reira, 2004). The DigitArq platform is the Portuguese

National archive system that uses well-established de-

scription standards, namely the ISAD(G) (General In-

ternational Standard Archival Description) (Interna-

tional Council on Archives, 2011) and ISAAR(CPF)

(International Standard Archival Authority Record

for Corporate Bodies, Persons and Families) (Vitali,

2004) with a hierarchical structure adapted to the na-

https://orcid.org/0000-0003-3744-2980

https://orcid.org/0000-0003-2370-3019

ture of archival assets.

To accomplish the migration process, an auto-

matic semantic migration prototype, based on Knowl-

edge Discovery, from Digital Archive metadata to

populate an ontology in CIDOC-CRM, was devel-

oped (Melo et al., 2022b). CIDOC-CRM (Con-

ceptual Reference Model) standard, an ontology de-

veloped for museums by the International Commit-

tee for Documentation (CIDOC) of the International

Council of Museums (ICOM) (Meghini and Doerr,

2018; ICOM/CIDOC, 2020), was used to build the

data model and description vocabularies (Koch et al.,

2019; Melo et al., 2020; Melo et al., 2022b). The se-

mantic migration process is supported by a set of de-

scription mapping rules that models the information

in CIDOC-CRM representation.

The resulting dataset is an OWL knowledge base

representation of the existing information in Digi-

tArq, where each DigitArq representation of meta-

data archives units has a scheme complying to

ISAD(G) (International Council on Archives, 2011)

and ISAAR (Vitali, 2004) recommendations. The

DigitArq information is organized according to a set

of ﬁelds and their values. Among this set of ﬁelds,

Varagnolo, D., Antas, G., Ramos, M., Amaral, S., Melo, D. and Rodrigues, I.

Evaluating and Exploring Text Fields Information Extraction into CIDOC-CRM.

DOI: 10.5220/0011550700003335

In Proceedings of the 14th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge Management (IC3K 2022) - Volume 2: KEOD, pages 177-184

ISBN: 978-989-758-614-9; ISSN: 2184-3228

177

there are some that present atomic values, such as the

’Reference code’, the ’Title’, or the ’Recipient’, that

do not require further interpretation, and the migra-

tion process was done by applying a predeﬁned set

of rules establishing the mapping between ISAD(G)

elements and CIDOC-CRM classes and properties.

There are other ﬁelds, such as ’Scope and content’

and Archival and Custodial History that are charac-

terized by having additional information describing

its unit, and it is in text format. These texts, usually,

have a structure that can be recognized, by using Nat-

ural Language Processing (NLP) tools, and giving as

output a feature value list that will be the input for the

additional ontology Population.

The process of extracting information from text

is not intended to extract all the information, but only

parts that are considered to be important, such as pass-

ports, baptisms, and births records, inventories due to

death, incorporation of documents between archives,

institutions, persons, and places involved in those ac-

tivities.

Methodologies to extract general information

from text into ontologies are presented in several

works, such as (Makki, 2017; di Buono et al.,

2014; Maynard et al., 2008; Petasis et al., 2011;

Lubani et al., 2019; Kordjamshidi and Moens, 2015;

Leshcheva and Begler, 2020). In particular, Onto-

Prima is a NLP-based Ontology Population system

that extracts instances of concepts and relations from

text to populate an ontology using NLP techniques.

The goal of this paper is to present an evaluation of

the automatic extraction of information about events

and entities, like persons, locations, dates, relations,

baptisms, births, etc; and the exploration of this infor-

mation to obtain a new organization of the archives

information.

The remainder of this paper is organized as follow.

Section 2 provides an overview about the events and

entities that are important to be extracted from text

ISAD(G) elements. The proposed approach for ex-

tracting events and entities from semi-structured text

is presented in Section 3. The evaluation of the extrac-

tion information process is detailed in Section 4. In

addition, the exploration of the CIDOC-CRM knowl-

edge base, through SPARQL queries, and the advan-

tages of the new information representation are also

presented. Finally, in Section 5, conclusions and fu-

ture work are drawn.

2 EXTRACTION OVERVIEW

The events and entities to be extracted from the doc-

uments text ISAD(G) elements depends on the type

Figure 1: Architecture of the Events and Entities Ontology

Population.

of the information provided by those documents. It

is possible to identify various types, such as passports

and baptisms records, births, inventories due to death,

incorporation of documents in archives, institutions,

persons, dates, and places involved in those activities.

The CIDOC-CRM ontology representation of this in-

formation and the corresponding mapping rules are

deﬁned manually and are presented in Section 3.2.

The type of the information, conveyed by a text

linked to a document, is determined by using an auto-

matic text classiﬁer. After the classiﬁcation, a proper

extraction process is applied. Some of these written

texts follow a semi-structured scheme, allowing the

deﬁnition of a methodology for extracting informa-

tion based on this scheme. The GATE

(General Ar-

chitecture for Text Engineering) environment is used

in this context, see Section 3 for more details.

Figure 1 presents the architecture of the proposed

approach that has 3 phases. The ﬁrst one runs a classi-

ﬁer over the national archives information OWL rep-

resentation. In the second phase, for each classiﬁed

text linked to a document (E

Document), the text

and the corresponding document reference code are

sent to an information extraction process. Finally, the

information extraction process extracts a set of rela-

tions, which, using the OWL rewriting rules (map-

ping rules), will represent in the CIDOC-CRM the ex-

tracted information linked to the document associated

with the target text of the extraction.

3 EXTRACTION PROCESS

The information extraction process from semi-

structured text is deﬁned using GATE, a Java suite

of tools to perform natural language processing tasks

over corpus. The tasks are managed by applica-

tions that include several language processing re-

sources. The main application is ANNIE (A Nearly-

New Information Extraction System), which is a set

https://gate.ac.uk/

KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development

178

Gazetteers

POS

Tagger

Transducer

Lists JAPE Rules

Semi-Structured

Text

ANNIE Processing

Rewriting Rules

Tokenizer

OrthoMatcher

Figure 2: Architecture of the Extraction Process from Semi-structured Text.

of modules comprising a tokenizer, a gazetteer, a sen-

tence splitter, a part-of-speech tagger, a named enti-

ties transducer, and a coreference tagger. Figure 2

presents the architecture of the information extraction

process from semi-structured text. Despite the fact

that GATE does not support (natively) the Portuguese

language, the test results obtained in the information

extraction tasks are very satisfactory and promising.

Each information extraction task is implemented

by deﬁning new rules for two modules:

The Gazetteer, originally, deﬁned as a geographi-

cal dictionary or a directory used in conjunction

with a map or atlas, are interpreted, in this work,

as entity dictionaries used in the Named Entity

Recognition task. In ANNIE application, these

entity dictionaries help to tag different words in

the texts.

The Named Entity Transducer is a technique of

Named Entity Extraction via Finite State Trans-

ducer, managed by rules. In ANNIE applica-

tion, it is possible to create and modify rules, us-

ing Jape

(Java Annotation Pattern Engine). A

Jape grammar consists of a set of phases, each of

which consists of a set of pattern/action rules. The

phases run sequentially and constitute a cascade

of ﬁnite state transducers over annotations. The

left-hand-side (LHS) of the rules consist of an an-

notation pattern description. The right-hand-side

(RHS) consists of annotation manipulation state-

ments. Annotations matched on the LHS of a rule

may be referred to on the RHS by means of labels

that are attached to pattern elements.

Consider, as an illustrative example, the text pre-

sented in the ’Scope and content’ of ’Jos

e Augusto

Teixeira’ record

, classiﬁed as a ’passport’. It is pos-

sible to identify a predeﬁned structure scheme in the

written text. For instance:

• The identiﬁcation of the parents of the applicant

{Filho de}: {father} e {mother}

https://gate.ac.uk/wiki/jape-repository/

http://digitarq.adbgc.arquivos.pt/details?id=1298186

• The occupation of the applicant

{Profiss

ao}: {occupation}

This type of structured text is suitable for deﬁning

rules that aid in the extraction of entities. The pattern,

as presented in this kind of text, is sufﬁcient to carry

out the extraction process only by applying the de-

ﬁned Jape rules, and there is no need to apply named

entity recognition techniques.

After the Gazetteer processing, the ﬁrst-degree

relative and the places that appears in the texts are

tagged. Then and after the processing of the Sentence

Splitter and the POS Tagger, the Jape rules are ap-

plied to extract the names of the parents, birthplaces,

parishes, and counties. Concerning the other text clas-

siﬁcation, specialized rules are deﬁned to extract the

information (in order to cover different patterns), that

have a priority system to be triggered.

As an example, the following Jape rules allows to

extract the father, the mother, and the occupation of

the applicant, from text classiﬁed as passport:

Rule: Parents Priority: 100

({Token.string ==˜ "[Ff]ilho"}

{Token.string == "de"} {Token.string == ":"}):start

({Person.kind == fullName}):father

({Token.string == "e"}|({Token.string == "e"}

{Token.string == "de"})):and

({Person.kind == fullName}):mother

-->

:father.Recipient = {field= "Father", recipient = "1" },

:mother.Recipient = {field= "Mother", recipient = "1" }

Rule: Profession Priority: 100

({Token.string ==˜ "[Pp]rofiss

ao"}

{Token.string == ":"}):start

(({Token.category == NN}|

{Token.category == NNP})+):occupation

-->

:occupation.Recipient = {field= "Occupation", recipient = "1" }

Finally, the output of each extraction task is a

XML ﬁle with all the information extracted. This ﬁle

is the input of the ’Rewriting Rules’ module that will

update the knowledge base with the new individuals,

class instances, and with the properties that link those

individuals. The rewriting rules represent the infor-

mation extracted in CIDOC-CRM. These new indi-

viduals are also linked to the document where they

were mentioned.

Evaluating and Exploring Text Fields Information Extraction into CIDOC-CRM

179

3.1 Passports Requisitions

Consider the DigitArq record of the passport requisi-

tion of ’Jos

e Augusto Teixeira’.The ISAD(G) ’Scope

and content’ element from the passport record con-

tains information about the applicant, such as parents

names, age, occupation, literacy, marital status, and

also the travel destination. The written information

contained in the ’Scope and content’ has an inten-

tional scheme in mind, with the aim of facilitating the

recording of information by archivists, as well as by

those who wish to consult it.

The information that can be extracted from the

ISAD(G) ’Scope and content’ element of passports

records:

Recipient - Name of the person that requested the

passport.

Father Name - Name of the father of the recipient.

Mother Name - Name of the mother of the recipient.

Place of Birth - Local, Parish and County Names in

Portugal.

Age - Age of the recipient

Literacy - Normally the annotation is: can write or

cannot write

Marital Status - The recipient marital state: mar-

ried, single, widow or divorced

Occupation - The recipient occupation, such as

farmer, domestic, mason, tailor, student, etc.

Place of Destination - A country or a country and a

town where the recipient is going to.

Passport Emission Date - The ISAD(G) ’date

range’ element of the requisition document.

3.2 CIDOC-CRM Representation

Each text classiﬁcation type has a set of mapping rules

assigned to it, which allows to represent the infor-

mation extracted in CIDOC-CRM. For better under-

standing, consider in particular the ’passport’ type of

information and the example previously presented in

Section 3.1.

The representation of a ’passport’ record in

CIDOC-CRM is presented in Figure 3. A ’passport’

requisition is represented as a CIDOC-CRM activity

activity) linked to the document representation by

property ’P

129

is about’. The ’E

activity’ has type

’passport’ and can have: a date, the applicant person

that is the document recipient, his parents, the destiny

place, and other information in the context of a Dig-

itArq passport record, see Section 3.1 for more de-

tails. In a passport record description the birth is also

implicitly described, when persons are referred and

the relative relations between them. The birth is rep-

resented by the CIDOC-CRM class birth (E

Birth)

linked to the document by the property ’P

refers to’.

The birth class has properties to represent the parents,

’P

by mother’ and ’P

by father’. New birth events

are used to represent the birth of the parents. All these

births are linked to the document with the text de-

scription by property ’P

refers to’.

This representation is automatically generated us-

ing the information extracted from the ’Scope and

content’ text, and from the document reference code

and title.

4 EVALUATION

The evaluation of the different extraction processes,

such as baptisms or passports descriptions, is done in-

dividually for each process topic considered. There-

fore, for each process classiﬁcation type, a set of

records was randomly selected from DigitArq and

with information accordingly classiﬁed with the pro-

cess topic. A dataset, a CSV ﬁle, is manually built

with the information of each set of records. 800

records are used for each topic. Another dataset

is build containing, for each set of records chosen,

the information represented in CIDOC-CRM and re-

trieved through SPARQL queries, The retrieved infor-

mation is saved in a CSV ﬁle with the same structure

of the dataset.

Dataset. A dataset for a speciﬁc topic contains the

information that the researcher considers to be ade-

quate to extract from the text. The dataset ﬁle is or-

ganized by a set of columns with the values of each

record in each row. Each topic is independently and

individually worked by 3 researchers. When the re-

searchers ﬁnish a topic, a ﬁnal dataset for the topic

is obtained by combining the information of the 3

datasets produced by them. The information pre-

sented in the ISAD(G) ’Scope and content’ element

may depend on how the Archives are used (Portu-

gal has one Archive for each District and there are

other institutions with their own archives, each one of

them following their own orientations on how to dis-

play and update the digital information). For instance,

in some archives the ’literacy’ is always displayed in

the passport records, but there are others that do not

include this information.

Evaluation Methodology and Results. The evalu-

ation uses as metrics the precision and the recall. In

each CIDOC-CRM dataset row, the non-empty value

KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development

180

Figure 3: CIDOC-CRM Representation of a Passport Requisition Activity.

of a ﬁeld (column) is correct if it matches the row

and column value of the manually built dataset. In

each CIDOC-CRM dataset row, the non-empty value

of a ﬁeld (column) is incorrect if it does not match the

row and column value of the manually built dataset.

Therefore, the Precision and Recall formulas are as

follows.

Precision =

number of correct values

number of correct values + number of incorrect values

(1)

Recall =

number of correct values

number of non-empty values in the Manual dataset

(2)

The Precision measures the quality of the values

extracted and the Recall measures the quantity of the

information extracted.

CIDOC-CRM dataset annotation provides the in-

formation needed to reﬁne information extraction

processes, allowing developers to look for error pat-

terns in natural language processing rules. New ver-

sions of the extraction processes can be evaluated in

the manually built datasets to test their adequacy.

These information extraction specialised pro-

cesses are still under development. New versions to

improve precision and recall or new topics that need

manually created datasets, new CIDOC-CRM repre-

sentations, and new Jape rules are being built. How-

ever, preliminary evaluation of the process for bap-

tisms and passports descriptions can achieve a preci-

sion of 98% and recall of 99%, when extracting dates,

entities, birth events, baptism activities or passport

requisition activities, with the correct roles assigned

to the persons who participated in those events, as

well as places.

4.1 Information Exploration

The information extracted from the semi-structured

text ﬁelds of the Portuguese Archives and represented

in CIDOC-CRM ontology enables new and different

content searches and visualizations of the information

taking into account the archives users needs (Hen-

nicke, 2013; Alma’aitah et al., 2020; Marlet et al.,

2019; Francart, ).

One example are the ’locals’ extracted, i.e., places

that are linked to activities or entities in some of these

information extraction processes, such as the pass-

ports requisitions, marriages, divorces or testaments.

In the archival records of these type of documents

the ISAD(G) ’Scope and content’ element allows the

recognition of the birth place of the document recip-

ient, and eventually the birth places of other persons

mentioned in the document, such as the parents of the

recipient. In particular and related to places:

Passport - describes the birth place of the passport

receipt.

Marriage - describes the birth place of the groom

and the bride as well as of their parents.

Divorce - describes the birth place of the ex-groom

and the ex-bride as well as of their parents.

Testament - describes the residence place of who

makes the testament.

After the extraction process applied over all these

kind of documents and the updating of the CIDOC-

CRM ontology population with the events and entities

extracted, it is possible to obtain some historical infor-

mation on the Portuguese administrative organization

Evaluating and Exploring Text Fields Information Extraction into CIDOC-CRM

181

for a year or a time period. For each processed doc-

ument, the ’Local’ is linked to the document repre-

sentation, ’E31 Document’, using the CIDOC-CRM

property ’P

refers to’. This property is inferred from

the facts extracted and the set of Semantic Web Rule

Language (SWRL)

rules such as the following ones.

R1: erlangen-crm:P01_has_domain(?pc129, ?doc)

ˆ erlangen-crm:P02_has_range(?pc129, ?person)

ˆ erlangen-crm:P129.1_has_type(?pc129, ’Recipient’)

-> erlangen-crm:P67_refers_to(?doc, ?person)

R2: erlangen-crm:P67_refers_to(?doc, ?person)

ˆ erlangen-crm:P96_brougth_into_life(?person, ?birth)

ˆ erlangen-crm:P7_took_place_at(?birth,?local)

-> erlangen-crm:P67_refers_to(?doc, ?local)

ˆ erlangen-crm:P67_refers_to(?doc, ?birth)

Rule R1 infers that all the persons that are recip-

ients from a document are referred by the document.

Rule R2 infers that if a person referred by a document

has a birth, then the birth and the place of birth are

referred by the document.

These rules are applied to the representations ob-

tained from speciﬁc kind of documents, such as pass-

port requisitions, baptisms, and other events (E7 Ac-

tivity), during the information extraction process from

the ISAD(G) ’Scope and content’ element.

Dates can be obtained from text descriptions or

from the ISAD(G) ’Date Range’ element, which

means that the archival material was produced in the

value of this ﬁeld, usually a date range.

In the archives modelling in CIDOC-CRM(Melo

et al., 2022b), every ’E

Document’ is linked to an

object (’E

Human Made Object) that may have a

production event (’E

Production’) linked to it. The

production event may have the date of production of

the archival material. The production event links this

date in CIDOC-CRM by the property ’P

108i

was pro-

duced by’ and the date is represented by an individual

of ’E

Time-span’ that can be qualiﬁed by begin and

end date.

As an example of the type of search that it is pos-

sible to achieve with this new representation of the

information, consider the following cases.

Query 1: What are the locals and their parishes lo-

cated in the county ’Braganc¸a’,

between 1900

and 1910?

Figure 4 presents this question expressed as an

SPARQL query diagram, based on the CIDOC-

CRM representation of the events, such as pass-

ports requisitions (see Figures 3). This dia-

gram can be straight forward translated into the

SPARQL query presented as follows.

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>

PREFIX owl: <http://www.w3.org/2002/07/owl#>

PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

https://www.w3.org/Submission/SWRL/

Braganc¸a is a Portuguese town.

PREFIX xsd: <http://www.w3.org/2001/XMLSchema#>

PREFIX cidoc: <http://erlangen-crm.org/200717/>

SELECT ?local ?parish WHERE{

?doc cidoc:P70_documents ?hmObj .

?hmObj cidoc:P108i_was_produced_by ?prod .

?prod cidoc:P4_has_time-span ?t1 .

?t1 cidoc:P79_beginning_is_qualified_by ?ti .

?t1 cidoc:P80_end_is_qualified_by ?tf .

?ti rdfs:label ?ti_datetime .

?tf rdfs:label ?tf_datetime .

BIND(year(xsd:dateTime(?ti_date_time)) AS ?yeari) .

BIND(year(xsd:dateTime(?tf_date_time)) AS ?yearf) .

filter(?yeari >= 1900 && 1910 >= ?yearf) .

?doc cidoc:P67_refers_to ?loc .

?loc cidoc:P89_falls_within ?local0 .

?loc cidoc:P89_falls_within ?local1 .

?loc cidoc:P89_falls_within ?local2 .

?pc0 cidoc:P01_has_domain ?local0 .

?pc0 cidoc:P1.1_has_type ?type0 .

filter(?type0 = "County Name") .

?pc0 cidoc:P02_has_range ?localApp0 .

?localApp0 rdfs:label ?local0 .

filter(?local0="Braganc¸a") .

?pc1 cidoc:P01_has_domain ?local1 .

?pc1 cidoc:P1.1_has_type ?type1 .

filter(?type0 = "Place Name") .

?pc1 cidoc:P02_has_range ?localApp1 .

?localApp1 rdfs:label ?local .

?pc2 cidoc:P01_has_domain ?local2 .

?pc2 cidoc:P1.1_has_type ?type2 .

filter(?type2 = "Parish Name") .

?pc2 cidoc:P02_has_range ?localApp2 .

?localApp2 rdfs:label ?parish }

This query evaluation lists every Local’s name

and its Parish’s name that occur in the text ﬁelds or

in other ISAD(G) elements, of the Portuguese Na-

tional Archives records, corresponding to archival

materials that were produced in the range from

1900 to 1910. Since this local’s names and its

parish’s names are the ones used in the speciﬁed

range, varying the range in this query, it is pos-

sible to obtain the nomenclature evolution in the

administrative organization of Portugal regions, a

result that can be useful for sociologists or his-

torians. A dataset, containing an excerpt of the

CIDOC-CRM representation of DigitArq records

from Braganc¸a District Archive, is available at

(Melo et al., 2022a). The dataset also includes

two SPARQL query examples to facilitate the ex-

ploration of the ontology.

At this moment, the ontology queries are per-

formed in SPARQL, a difﬁcult task even for ex-

perts. However, if graph query diagrams, like the

one shown in Figure 4 are constructed, the task of

translating it into a SPARQL query is facilitated.

KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development

182

Figure 4: Graph Representation of the Query 1.

Query 2: What is the number of children per couple,

between 1800 and 1850?

Figure 5: Graph Representation of the Query 2.

This query may be represented by the graph dia-

gram presented in Figure 5, that can be translated

into the following SPARQL query.

SELECT ?motherName ?fatherName

(count(?son) as ?numberSons)

WHERE{

?birth cidoc:P98_brought_into_life ?son .

?birth cidoc:P97_from_father ?father .

?father cidoc:P1_is_identified_by ?fatherName.

?birth cidoc:P96_by_mother ?mother .

?mother cidoc:P1_is_identified_by ?motherName.

?birth cidoc:P4_has_time-span ?t1 .

?t1 cidoc:P170i_time_is_defined_by ?t11 .

?t11 rdfs:label ?t1_datetime .

BIND(year(?t1_datetime) AS ?year).

filter(?year >= 1800 && 1850 >= ?year)

} group by ?motherName ?fatherName

Birth events are extracted of many types of

archival materials from the Portuguese Catholic

Church Archives, referring to baptisms and mar-

riages, or from the Portuguese Civil Adminis-

tration Archives, on passport requisitions, tes-

taments, marriages or divorces. The informa-

tion extracted from these type of documents is

used in the above query, integrating the content

of all Portuguese archives, including the catholic

church ones that normally are incorporated in

Civil archives.

5 CONCLUSIONS

A method for extracting information from ISAD(G)

elements, that contain semi-structured text descrip-

tions, was proposed in this paper. The method has 5

phases: the deﬁnition of the information to be rep-

resented, the deﬁnition of the rules to map the in-

formation into CIDOC-CRM representation, the Jape

rules necessary to process the text and extract the in-

tended information, a sample of 800 records with the

type of information, and the dataset manually built

for each type of information considered. These steps

allowed the deﬁnition and implementation of sev-

eral automatic information extraction processes that

can be both evaluated and populate the CIDOC-CRM

knowledge base with new linked events and entities

automatically.

With the information of the archival records rep-

resented in CIDOC-CRM, it is possible to retrieve in-

formation that was not easily available in the Digi-

tArq, the relational database, that only has public ac-

cess by a portal with simple and advanced search that

does not include SQL queries. With the Portuguese

archives information migrated into a CIDOC-CRM

knowledge base, the information can be searchable

using SPARQL or DLquery, allowing new visualisa-

tions of the archival records and the retrieval of in-

formation collected in different records from differ-

ent archives. Two examples of these queries were

presented that highlight the new types of retrieved in-

Evaluating and Exploring Text Fields Information Extraction into CIDOC-CRM

183

formation facilitated by the representation in CIDOC-

CRM. The exploration of the CIDOC-CRM informa-

tion is done in SPARQL or DLqueries, which can be

difﬁcult to use by researchers who are interested in

exploring the archives content. In a near future, this

project work will focus on a friendly interface capa-

ble of automatically generate the SPARQL queries to

CIDOC-CRM from user natural language queries.

ACKNOWLEDGEMENTS

This work is ﬁnanced by National Funds through

FCT - Foundation for Science and Technology

I.P., within the scope of the EPISA project -

DSAIPA/DS/0023/2018.

REFERENCES

Alma’aitah, W., Talib, A. Z., and Osman, M. A. (2020).

Opportunities and challenges in enhancing access to

metadata of cultural heritage collections: a survey. Ar-

tiﬁcial Intelligence Review, 53(5):3621–3646.

di Buono, M. P., Monteleone, M., and Elia, A. (2014).

How to populate ontologies. In M

etais, E., Roche,

M., and Teisseire, M., editors, Natural Language Pro-

cessing and Information Systems, pages 55–58, Cham.

Springer International Publishing.

Francart, T. Sparnatural - Javascript SPARQL query

builder. https://sparnatural.eu/. Accessed: 2022-09-

Hennicke, S. (2013). Representation of archival user

needs using cidoc crm. In Conference Proceedings

TPDL: International Conference on Theory and Prac-

tice of Digital Libraries, Selected Workshops, Valletta,

Malta.

ICOM/CIDOC (2020). Deﬁnition of the CIDOC Concep-

tual Reference Model. ICOM/CRM Special Interest

Group, 7.0.1 edition.

International Council on Archives (2011). ISAD(G): gen-

eral international standard archival description, Sec-

ond Edition. Springer Nature BV.

Koch, I., Freitas, N., Ribeiro, C., Lopes, C. T., and da Silva,

J. R. (2019). Knowledge graph implementation of

archival descriptions through cidoc-crm. In Doucet,

A., Isaac, A., Golub, K., Aalberg, T., and Jatowt, A.,

editors, Digital Libraries for Open Knowledge, pages

99–106, Cham. Springer International Publishing.

Kordjamshidi, P. and Moens, M.-F. (2015). Global machine

learning for spatial ontology population. Journal of

Web Semantics, 30:3–21.

Leshcheva, I. and Begler, A. (2020). A method of semi-

automated ontology population from multiple semi-

structured data sources. Journal of Information Sci-

ence, 0(0).

Lubani, M., Noah, S. A. M., and Mahmud, R. (2019). On-

tology population: Approaches and design aspects.

Journal of Information Science, 45(4):502–515.

Makki, J. (2017). Ontoprima: A prototype for automating

ontology population. International Journal of Web/Se-

mantic Technology (IJWesT), 8.

Marlet, O., Francart, T., Markhoff, B., and Rodier, X.

(2019). OpenArchaeo for Usable Semantic Interoper-

ability. In ODOCH 2019 @CAiSE 2019, Rome, Italy.

Maynard, D., Li, Y., and Peters, W. (2008). Nlp techniques

for term extraction and ontology population. In Pro-

ceedings of the 2008 Conference on Ontology Learn-

ing and Population: Bridging the Gap between Text

and Knowledge, page 107–127, NLD. IOS Press.

Meghini, C. and Doerr, M. (2018). A ﬁrst-order logic ex-

pression of the cidoc conceptual reference model. In-

ternational Journal of Metadata, Semantics and On-

tologies, 13(2):131–149.

Melo, D., Rodrigues, I. P., and Koch, I. (2020). Knowl-

edge discovery from isad, digital archive data, into

archonto, a cidoc-crm based linked model. In Pro-

ceedings of the 12th International Joint Conference

on Knowledge Discovery, KEOD - Volume 2, pages

197–204. INSTICC, SciTePress.

Melo, D., Rodrigues, I. P., and Varagnolo, D. (2022a). Por-

tuguese Examples (Semantic Migration), Mendeley

Data, V1. 10.17632/t2cx9stwfb.1.

Melo, D., Rodrigues, I. P., and Varagnolo, D. (2022b).

A strategy for archives metadata representation on

cidoc-crm and knowledge discovery. Semantic Web,

Pre-press(Pre-press):1–32.

Petasis, G., Karkaletsis, V., Paliouras, G., Krithara, A., and

Zavitsanos, E. (2011). Knowledge-Driven Multime-

dia Information Extraction and Ontology Evolution:

Bridging the Semantic Gap. Springer Berlin Heidel-

berg, Berlin, Heidelberg.

Ramalho, J. C. and Ferreira, J. C. (2004). Digitarq: cre-

ating and managing a digital archive. In Building

Digital Bridges: Linking Cultures, Commerce and

Science: 8th ICCC/IFIP International Conference on

Electronic Publishing held in Bras

ılia - ELPUB 2004,

Brazil, June, 2004.

Vitali, S. (2004). Authority control of creators and the

second edition of isaar (cpf), international standard

archival authority record for corporate bodies, per-

sons, and families. Cataloging & classiﬁcation quar-

terly, 38(3-4):185–199.

KEOD 2022 - 14th International Conference on Knowledge Engineering and Ontology Development

184