SIOP-LEGIS: THESAURUS FOR SELECTION AND

MANAGEMENT OF BRAZILIAN TREASURY DOMAIN

Stainam Brandão

, Tiago Silva

, Sergio Rodrigues

, Luis Araujo

, Daniel Silva

and Jano Souza

COPPE/UFRJ, Graduate School of Engineeringy, Rio de Janeiro, Brazil

Brazilian Department of the Treasury, Brasilia, Brazil

Keywords: Thesaurus evolution, e-Government, Knowledge representation.

Abstract: This work presents the approach used to select documents from the Brazilian Official Gazette for Federal

Treasury domain and manage the explicit knowledge through a thesaurus according to ISO Standard 2788.

The goal of the SIOP-LEGIS project joint with Department of Treasury is to facilitate search and retrieval

of documents semantic related through hierarquic terms defined in the Thesaurus. The paper also presents a

brief case study to demonstrate how the ontology domain evolution ocurrs in accordance with the

legislation, case law and administrative acts in constant changes.

1 INTRODUCTION

The Secretary of Federal Treasury, during the work,

needs information to suport decision making and

justify their actions legally. Likewise, access to

legislation, jurisprudence and administrative actions

is required from the Brazilian Official Gazette.

However, read and follow the jurisprudence takes

time and dedication. SIOP-LEGIS is a Knowledge

Organizational System that mines Brazilian Official

Gazette daily generating metadata for each

document related to Federal Treasury domain.

Nowadays, SIOP-LEGIS repository has a

collection of more than 80.000 indexed documents

since 2008 and that number increases daily.

However, only documents relevant to the domain are

selected for further validation of experts.

Selection of Documents

Initially, domain experts have developed a thesaurus

for the Federal Treasury with the main terms of

Federal Budget area, whereas the system

automatically selects the documents that would be

part of through the use of thesaurus. After, the expert

validates all of the selected documents, assigning

one of three statuses:

 Validated: document selected automatically and

validated by expert;

 Discarded: document automatically selected

and discarded by an expert, not being a

document of interest to Federal Budget area;

 Not selected: document not being selected

(either automatically or manually).

Validation of Documents

There is a experts team in Treasury Department to

perform the daily reading of the Federal Official

Gazette and all decisions of the Supreme Court. This

team also validate the documents selected and read

the documents not selected for the evaluation of the

methodology presented in this paper.

After validation, the document metadata created

automatically for each validated document will be

used as input for domain knowledge representation.

At this point we present our approach through

actions for knowledge representation and evolution,

described in the next section.

2 REPRESENTATION AND

EVOLUTION ACTIONS

Since the manual thesaurus construction requires a

huge human effort, the acquisition of which is

considered a bottleneck for the Knowledge

Management. In this paper, domain thesaurus

evolution involves enrichment through actions on

the vocabulary, which is representing the explicit

knowledge of Brazilian Official Gazettes to

semantically index them.

This section presents a dynamic and adaptive

representation through actions and rules that evolves

195

Brandão S., Silva T., Rodrigues S., Araujo L., Silva D. and Souza J..

SIOP-LEGIS: THESAURUS FOR SELECTION AND MANAGEMENT OF BRAZILIAN TREASURY DOMAIN.

DOI: 10.5220/0003660301950200

In Proceedings of the International Conference on Knowledge Management and Information Sharing (KMIS-2011), pages 195-200

ISBN: 978-989-8425-81-2

 2011 SCITEPRESS (Science and Technology Publications, Lda.)

domains thesaurus identifying events those demands

new configuration.

The indexing of documents in the domain

thesaurus uses an approach to document adherence

to each concept, through occurrence and proximity

of terms. This work monitors events related to the

quality metrics of domain ontologies and the

changes occurring on the documents collection

(Corpus) over time.

Our proposal handles domain thesaurus and these

only treat the documents belonging to the same

domain. There is no intention to identify the domain

of a document and extend the thesaurus to another

domain.

The next sub-sections present each action.

2.1 LDA / LSA

This action uses the topic modelling approach

through the Latent Dirichlet Allocation - LDA (X.

Wei e W. B. Croft, 2006) to identify topics for

Corpus representation and after, applies the Latent

Semantic Analysis - LSA (S. Cederberg e D.

Widdows, 2003) approach to identify the terms

correlated with higher values for latent semantic

with the topics. These Topics and correlated terms

are identified as candidates for the thesaurus

concepts.

Applicability

The representation of words and sentences meaning

captured by the LSA approach has been able to

simulate the recognition of vocabulary and

categorizing words, which make possible to

approximate the human judgments of similarity

between words, an aspect that is highly valued in the

study of speech processing.

Model

Usually, topics are small and mostly consist of a

single word (Watters D., 2009). Each topic

represents ideas of origin, purpose or description of

the documents content. In addition, the approach is

applied on the 'term-document" matrix

representation and applies the dimension reduction

technique based on Singular Value Decomposition

(SVD). In this moment, the documents and words

are mapped into a representation of the latent

semantic space, which is based on topics rather than

each word individually and therefore the space of

representation is much smaller than the original.

This control is temporal and every change in the

documents collection leads to recalculated the topics

and correlated terms to identify terms that have been

outstanding, emerging and deprecating, according

figure 1.

Figure 1: Mapping candidates terms: (1) mapping between

topics (T) and correlated terms (t), (2) mapping between

the thesaurus concepts © and topics (T) / correlated terms

(t).

Example

After ran the crawler, register and pre-processing

only the contents of the site of Brazilian Social

Security (

http://www.previdenciasocial.gov.br/conteudo

Dinamico.php?id=690

), the proposed approach

extracts and presents the topics and their respective

correlated terms with the highest values of latent

semantic for the experts. These terms correlated not

only try to explain the meaning of the topic within

the text but also serve as candidates to the concepts

of domain thesaurus.

Figure 2: One of the Text pre-processing used for

extracting topics.

Figure 3: Two Topics with their correlated terms in

Portuguese language: ‘Comportamento’ and ‘Flutuações’.

2.2 Concept Fragmentation

This event occurs when the thesaurus concepts have

excess pointers to the documents indicating the need

KMIS 2011 - International Conference on Knowledge Management and Information Sharing

196

for concept specialization with the identification of

subclass in the collection of documents being

referenced.

This overhead shows the importance of the

concept for the domain and the possibility of new

related concepts in the documents.

Applicability

The evolution must be based on clear rules to avoid

inconsistency. And the Fragmentation acts on a

concept considered important for the domain

ontology due to the large number of documents

indexed, which allows the identification of

subclasses in documents.

This action has two strategies:

· Complete: if all subclasses cover all documents

indexed, then the concept will be removed;

· Incomplete: If the subclasses do not cover all

documents referenced, then the concept is retained

in the domain ontology.

Model

The Fragmentation action uses guard expressions

associated with the following metrics: Importance

(Quantifies instances that belong to a concept in a

sub-tree) and Class Richness (distribution of

instances between concepts).

When some of these metrics achieve a value that

triggers at least one of the guard expressions, the

Fragmentation action is performed:

1. Identify the concept that triggered the

guard expression;

2. Select the Terms Correlated with greater

similarity with the concept (identified in step 1) in

the documents indexed;

3. Analysis the completeness of the

Correlated of Terms:

a. If the subclasses are complete, then the

concept identified in step 1 is removed;

Figure 4: Strategy: Complete.

b. If the subclasses are incomplete, so the

concept identified in step 1 is kept in the model.

Figure 5: Strategy: Incomplete.

4. Redistribution of the documents indexed

among the concepts involved. Using the approach to

document adherence to each concept, through

occurrence and proximity of terms.

Example

The figure 6 shows a hypothetical concept 'Pensão'

(‘pension’, in English) which indexes a collection of

documents and has no subclasses. Whilst figure 7,

shows the identification of ten subclasses that

represent faithfully the documents collection

previously referenced by the term 'Pensão' deleted.

Figure 6: Concept ‘Pensão’ has no subclass.

Figure 7: Fragmentation action: ten subclasses identified

and concept ‘Pensão’ deleted.

2.3 Abstraction

This action transforms the domain ontology in a

faithful representation of the Corpus. However, the

knowledge already represented cannot be lost,

because each application uses the knowledge that

supports your desired area. So the idea is to use

semantic relativity in the model in an accordance of

the necessity of the application. Therefore, a

versioning control is used.

Applicability

The applicability is related to ability to provide

thesaurus abstractions (views) for applications that

operate in different areas of the same domains

represented by the ontology.

SIOP-LEGIS: THESAURUS FOR SELECTION AND MANAGEMENT OF BRAZILIAN TREASURY DOMAIN

197

Model

The model uses the Corpus to identify the thesaurus

concepts missing as well as highlights those has its

co-occurrence and frequency decreased overtime.

For this, future approaches will be used TF-IDF and

LDA / LSA.

Figure 8: Abstraction per sub-domain.

Example

Within the Federal Treasury domain, a thesaurus

was developed initially for the Federal Budget area,

due to a system for registering, searching and

retrieving documents from the Brazilian Official

Gazette. However, with the initial success of the

proposal, two other distinct areas of Federal

Treasury domain have expressed interest in using the

system: Human Resources Secretary (SRH) and

Shareholders' Union Secretary (SHF). Thus, this

action provides conditions from a single thesaurus

providing three abstractions (views) of the domain to

different areas that may have common concepts as

distinct concepts.

3 SIOP-LEGIS

The Knowledge Organizational System (SIOP-

LEGIS) is a project that aims to provide knowledge

management for legislative domain through

changeable representation, which deal trends in the

law. Its architecture is presented in Figure 9.

Figure 9: Initial Architecture.

Natural Language Processing

Each known document from Brazilian Official

Gazettes has a type: Order, Law, Internal Program,

etc. Since the relationships initially defined between

them are: revoke, alter, establish, regulate and

deploy and create. All of those are modelled in an

RDF ontology for not to be considered during the

natural language processing. Likewise stopwords

and list of known entities that contains authorities

(people and organizations), technical terms, dates

and values that are not of interest for representation

in the thesaurus

4 CASE STUDY

This section presents a brief case study to

demonstrate how the architecture works. The

example used in this work is adapted from Secretary

of Federal Treasury responsible for control and

oversight of federal spending in accordance with the

legislation, case law and administrative acts.

For this, the SIOP-LEGIS system was used for semi-

automatic selection and indexing of documents from

Brazilian Official Gazettes that deals with the

Federal Treasury domain.

Figure 10: This caption shows de topics extracted from a

validated document.

From the domain thesaurus of the Department of

Treasury, the system automatically pre-selects the

documents that would be part of this domain. After,

experts validate the pre-selected documents that

acquire one of three statuses:

KMIS 2011 - International Conference on Knowledge Management and Information Sharing

198

(1) Validated: document selected automatically and

validated by an expert;

(2) Discarded: document automatically selected and

discarded by expert not being a document of

interest;

(3) Not selected: document not being selected

automatically.

Each validated document was indexed on thesaurus

terms and used by the LDA / LSA action for

identification of new terms and relations. Figure 10

and 11 show two documents validated with the

terms identified by LDA / LSA for the domain

thesaurus evolution.

Figure 11: This caption shows another topics extracted

from a validated document that will enrich the thesaurus.

The methodology generates a new version of

thesaurus with the inclusion of the following

knowledge represented.

Figure 12: This caption show de thesaurus evolution from

the two documents previously captured, according ISO

2788. TG: concepts with wider connotation; TE: concepts

with more specific connotation; TR: concepts with more

specific connotation.

Currently, the domain thesaurus is used as artefact

on the recommendation mechanism for search and

retrieval in the same system.

Figure 13: Recommendation mechanism for search.

5 CONCLUSIONS AND FUTURE

WORKS

This paper presents a methodology applied by

Secretary of Federal Budget for knowledge

representation and management of Brazilian Official

Gazette for Federal Treasury domain. A domain

thesaurus is used to represent a changeable domain

with a daily inclusion of new Gazette with

legislation, jurisprudence and administrative actions.

One future work is the development of a

Context-Sensitive Search based on the content

documents analysed, with identification of resources

and relationship previously specified on domain and

represented on RDF ontology.

Another future work is allow navigation through

documents by these RDF resources, deeming the

correlation of documents already defined with the

Vector Model (TF-IDF) approach from (Lucene,

2009) to calculate the similarity between the

documents.

A future target is the integration with Formal

Ontology to establish formal meanings for domain

vocabulary allowing axiomatization and integration

of domain ontologies from different sources. We

notice that domain ontologies consist of specialized

terminology and a particular vision of reality. But

the meaning remains dependent on the context. The

use of Formal Ontology to integrate the domain

ontology can promote the reuse, integration and

management through the construction of ontologies

from different countries about the same domain.

ACKNOWLEDGEMENTS

We thank CAPES, CNPq and Department of

Treasury for supporting this work.

SIOP-LEGIS: THESAURUS FOR SELECTION AND MANAGEMENT OF BRAZILIAN TREASURY DOMAIN

199

REFERENCES

A. Maedche, B. Motik, L. Stojanovic, R. Studer, e R.

Volz, “An infrastructure for searching, reusing and

evolving distributed ontologies,” WWW, 2003, pp.

439–448.

B. Omelayenko, “Learning of Ontologies for the Web: the

Analysis of Existent Approaches,” Proceedings of the

International Workshop on Web Dynamics, held in

conj. with the 8th International Conference on

Database Theory (ICDT’01), London, UK, 2001.

Clips, “http://clipsrules.sourceforge.net/” Acessed in

august 22th 2009.

G. Guizzardi, “Ontological foundations for structural

conceptual models,” 2005.

H. Alani e C. Brewster, “Ontology ranking based on the

analysis of concept structures,” Proceedings of the 3rd

international conference on Knowledge capture,

Banff, Alberta, Canada: ACM, 2005, pp. 51-58.

Jaccard, “http://sourceforge.net/projects/simmetrics/”

Acessed in august 22th 2009.

Jess, “http://www.jessrules.com/” Acessed in august 22th

2009.

K. Beyer, J. Goldstein, R. Ramakrishnan, e U. Shaft,

“When Is "Nearest Neighbor" Meaningful?,” In Int.

Conf. on Database Theory, 1999, pp. 217--235.

M. W. Geert, G., “The Ontological Foundation of REA

Enterprise Information Systems,” Ago. 2000.

N. Guarino, “Formal Ontology and Information Systems,”

1998.

M. Shaw e D. Garlan, Software Architecture: Perspectives

on an Emerging Discipline, Prentice Hall, 1996.

N. Russell, A. ter Hofstede, D. Edmond, e W. van der

Aalst, “Workflow Data Patterns: Identification,

Representation and Tool Support,” Conceptual

Modeling – ER 2005, 2005, pp. 353-368.

N. Russell, Arthur, W. van der Aalst, e N. Mulyar,

Workflow Control-Flow Patterns: A Revised View,

2006.

S. A. Macskassy, A. Banerjee, B. D. Davison, e H. Hirsh,

“Human Performance on Clustering Web Pages: A

Preliminary Study,” KDD, 1998, pp. 264-268.

T. Berners-Lee, J. Hendler, e O. Lassila, “The Semantic

Web A new form of Web content that is meaningful to

computers will unleash a revolution of new

possibilities,” 2001.

T. R. Gruber, “A translation approach to portable ontology

specifications,” Knowl. Acquis., vol. 5, 1993, pp. 199

Gomez-Perez, A., Knowledge sharing and reuse, 1998. In

J. Liebowitz, ed., The Handbook of Applied Expert

Systems, CRC Press.

Lucene, “http://lucene.apache.org/” Acessed in november

12th 2009.

Watters D., “Meaningful Clouds: Towards a novel

interface for document visualization”. Disponível em:

<http://danwatters.com/documents/CloudMine_dwatte

rs.pdf>. Acesso em 09/09/2010, 2009.

S. Cederberg e D. Widdows, “Using LSA and noun

coordination information to improve the precision and

recall of automatic hyponymy extraction”, in

Proceedings of the seventh conference on Natural

language learning at HLT-NAACL 2003 - Edmonton,

Canada, 2003, pp. 111-118.

X. Wei e W. B. Croft, “LDA-based document models for

ad-hoc retrieval”, in Proceedings of the 29th annual

international ACM SIGIR conference on Research and

development in information retrieval, New York, NY,

USA, 2006, p. 178–185.

KMIS 2011 - International Conference on Knowledge Management and Information Sharing

200