INTERPRETATION AND RECOMMENDATION TASKS

SUPPORTED BY CERES SYSTEM

Cristina Paludo Santos and Denilson Rodrigues da Silva

Departament of Engineering and Computer Science, URI University, Santo Ângelo, Brazil

Keywords: Knowledge Representation, Ontology, Soil Chemical Analysis, Agricultural Domain.

Abstract: This article describes the interpretation and recommendation modules of the Ceres system, an Expert

System that supports the interpretation of soil samples, recommendation for lime, fertilizer and data

management using resources from expert system and database technologies. Among the requirements of the

application domain, it is the information analysis resulting from the interpretation process, which has the

purpose of finding the causes of productivity variation. This analysis makes possible human interaction in

the sense of preserving proprieties that favor the developing of vegetables. In this way, the Ceres system has

been proposed to assist experts on possible actions to be implemented to improve soil quality. The system is

compound by several modules that implement tools to analyse the different soil properties. The focus of this

paper is to analyze soil chemical aspects and the analysis tasks related to them. The modules proposed were

conceived with a knowledge base formed by ontology, an inference engine and an interface for external

access.

1 INTRODUCTION

Precision Agriculture is a set of procedures that aims

to manage a productive field, meter by meter,

considering the different characteristics of each

piece of the property (Roza 2000). In order to do so,

several techniques are applied, such as productivity

maps (Molin 2000), geological information systems

(Fileto 2005), and analysis for fertility, with the

objective of obtaining precise information about the

conditions of the property in the period before the

production cycle.

More specifically, the analysis for fertility

involves the study of the causes of the variation of

productivity and the development of vegetables in

the agricultural fields. For this, different factors

involved in the development of plants are examined

individually. Each study results in different sets of

analysis to be interpreted by specialists in the field.

In this way, the specialist is able to make decisions

related to the necessary actions for maintaining the

properties of soil fertility, to increase the

productivity and the environment preservation.

In this context, is proposed the Ceres System -

an expert system that supports the activities of

interpretation of the soil proprieties and recommends

the use of fertilizing. It integrates the benefits of the

database and expert system technologies to provide

facilities to information management by making

storage, access, data and knowledge application

easier in this domain.

The system is compound by 5 expert modules,

being presented in this paper the interpretation and

recommendation modules related to the soil’s

chemical aspects.

The next sections are organized as follows.

Section 2 describes the main characteristics of the

Ceres System and their architecture. Section 3

presents the knowledge representation emphasizing

the knowledge modeling involved in the

interpretation and recommendation modules and the

ontology implemented. Section 4 presents the

prototype developed. The main conclusions of the

work are presented in section 5.

2 THE CERES SYSTEM

Ceres is an expert system that automatizes the

activities of interpretation of the soil proprieties and

recommends the use of fertilizing, when necessary.

In this manner, the system receives soil’s

descriptions and makes interpretation according to

the knowledge acquired and formally represented in

the correspondent modules. Therefore, the inference

resultant of the interpretation process is used by the

recommendation module to indicate the actions to be

taken for the maintenance of soil fertility’s

464

Santos C. and da Silva D. (2009).

INTERPRETATION AND RECOMMENDATION TASKS SUPPORTED BY CERES SYSTEM.

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 464-467

DOI: 10.5220/0002309304640467

 SciTePress

properties. The data and information are preserved

to queries and used in the future.

The Ceres system is compound by three main

components: a symbolic processing; a database

system and a visual interface. The symbolic system

is the core intelligent of the system and consequently

is the main component of the architecture. The

function of this component is decomposed in five

modules, as presented in Figure 1. The

modularization of this layer allows the

implementation of specific knowledge to develop

the tasks related to each type of analysis.

The database system component stores all

descriptions managed by the system and the

structure of the knowledge base of the application

domain. This component is supported by a RDBMS

(Relational Data Base Management System) that

controls the access of the sample descriptions and

implements the integrity constraint of the model.

There is no domain knowledge in the system which

is not managed by the database module.

The third component – the user interface

supports the new soil sample description that will be

stored in the database system. The integrated

environment of multiple windows offered by the

system to the user allows the standardization and

facilitates the process of soil chemical analysis.

Figure 1: Ceres System Architecture.

The interpretation and recommendation modules

described were included in the symbolic system and

are represented in Figure 1 by dotted lines.

The system’s architecture supports expert system

and database functionality with high flexibility,

providing a reliable environment to help specialists

in reasoning with big amount of data. Besides, the

system is able to represent the knowledge preserving

the necessary semantic richness of the domain and

offers a user-friendly environment to accomplish the

complex task of soil analysis.

3 THE KNOWLEDGE

REPRESENTATION

The knowledge acquisition was realized using the

following techniques: (a) Interviews with specialist

more than ten-year-experience, (b) bibliographical

study of the liming and fertilization manual

(Sociedade Brasileira de Ciência do Solo 2004) and,

soil chemical analysis.

The studies, analysis and interviews enabled the

development of the ONIACHES - an ontology that

supports the systematization of the interpretation and

recommendation processes related to soil chemical

analysis task. The ontology created provides a

reference domain model that human and software

can refer to for several different purposes. A more

detailed description about the ontology ONIACHES

can be found in (SANTOS 2007).

To construct the ONIACHES it was used the

METHONTOLOGY (Gómez-Pérez 2003) and

CommonKADS (Schreiber 2000) methodologies.

METHONTOLOGY was adopted since it is more

expressive, predicts the integration of ontology,

provides good tools where it can be applied and it is

recommended by the FIPA (Foundation for

Intelligent Physical Agents). Inference structures are

used in order to describe the reasoning process,

according to CommonKADS methodology, which

allows the representation in knowledge level of the

relationship between rules for execution of the task.

In addition, to further specify the inferences realized

in knowledge level, knowledge graphs were

modelled (Leão 1990).

In summary, the following activities were

developed: (1) knowledge acquisition; (2)

elaboration of specification document; (3)

elaboration of terms glossary; (4) concepts

taxonomy modeling; (5) concepts dictionary

elaboration; (6) detailed definition of the

relationship (relation’s diagram and relation’s

dictionary); (7) detailed definition of the attributes

(attribute’s dictionary); (8) definition of instances

(instance’s dictionary); (9) inference structures

modeling; (10) definition of axioms and rules;

resulting in 17 elaborated documents.

The elaboration of these documents allows the

verification of the complexity and the real dimension

of the ontology. It also provided detailed and well

AnalyticalInterface

UserInterface

SampleDescription

Information

Additional

Help

Interface

SymbolicSystem

SystemOb

ects

JDBCInterface

Database

RelationalDataBaseManagementSystem

Knowledge

ChemicalAnal.

Interpretation

Module

PhysicalAnal.

Interpretation

Module

BiologicalAnal.

Interpretation

Module

FoliarAnalysis

Interpretation

Module

Recomendation

Module

INTERPRETATION AND RECOMMENDATION TASKS SUPPORTED BY CERES SYSTEM

465

documented models contributing to a better

understanding of the area.

3.1 The Interpretation Module

The input data for the interpretation module are

derived from laboratory analysis performed on soil

samples. The laboratory analysis allows extract

soil’s chemical properties, such as PH, content of

clay, boron, copper, magnesium, potassium,

phosphorus and others. These properties should be

correlated to allow its interpretation.

To specify in detail the inferences realized in

knowledge level, 24 graphs were modelled, each one

representing different relations between the elements

involved in the interpretation process.

The knowledge base is formed by 125 instances

of concepts of the ontology to interpret the 15

attributes related to the interpretation task. The

inference employed in the process was represented

by rules definition. Ten rules were defined for the

representation of the inference inherent to the

interpretation process. In the documentation

generated, each rule has a name, a formal

expression, and the description of the variables used.

3.2 The Recommendation Module

The recommendation process supported by the Ceres

system involves the tasks of recommendation of

fertilizers and recommendation of liming. In order to

be possible to incorporate the recommendation

process in the system, it was necessary to extend the

ONIACHES integrating new concepts and giving

origin to a new version of the ontology.

As a whole, 634 new instances of concepts

related to recommendation process were created,

being 445 instances used to represent the task of

recommendation of fertilizers and 189 instances to

represent the task of liming recommendation. All the

instances are documented, including, to each one of

them, attributes, concepts and associated values.

The inference employed to the recommendation

process was represented by the definition of 27

rules. These rules handle the instances of concepts of

the ontology.

To establish the relationships between the

concepts involved in the ontology, a relation’s

diagram has been elaborated. It allows the

visualization of the existing independence between

the interpretation module and the recommendation

module. That is, the interpretation module can be

used without any relation to the recommendation

module. However, this is not the case of the

recommendation module, which has dependence of

the data come from the interpretation module. The

diagram has been elaborated with a schematic

approach, allowing a clearer vision of the concepts

and their roles in the application domain.

3.3 The Ontology Implementation

It was possible to instance the knowledge base from

the modelled and documented knowledge. The base

has been constructed with the support of the Protegè

Edition Tool (Noy 2003) version 3.3, which supports

the methodology chosen, allows automatic

documentation, facilitates the integration of

ontology and gives support to the importation and

exportation in several formats, as OWL (Web

Ontology Language) and RDF Schema. The use of

Protégé made it possible the edition of concepts

taxonomy, relationships and attributes of the

ontology and the instantiation of 759 concepts

related to interpretation and recommendation

processes.

To realize the inference process, Algernon

(Hewett 2004) has been used, an inference engine

which uses a rules-based language capable of

manipulating the knowledge bases instantiated in the

Protégé. Algernon was chosen for the following

reasons: (i) it is an open source tool; (ii) it has good

documentation available; (iii) it supports the

backward chaining and forward chaining methods;

(iv) it is developed in the same language that Protégé

and, (v) it has maturity enough to be used and

encapsulated in an application.

The characteristics described above about

Protegé and Algernon tools enabled the development

of software that encapsulates the complete structure

of symbolic data processing in an expert system able

to realize the task in accordance with the knowledge

modeled in the ontology.

4 THE PROTOTYPE

DEVELOPED

With the aim of modelling the system’s prototype, it

was used UML language, which has enough

representativeness and wide documentation

available. The use case diagram, component diagram

and class diagram were elaborated to document the

system’s prototype. The prototype was implemented

in Java language allowing the encapsulation of

Protégé and Algernon technologies in a single

application. The component’s structure supported by

prototype is presented in Figure 2.

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

466

Figure 2: Component’s Structure.

After the implementation of the specialist

system, usability and functionability tests were

performed in order to validate the prototype. In this

sense, tests evidenced, satisfactorily, validity and

reliability in the tool use.

5 CONCLUSIONS

The use of geoprocessing techniques and Global

Positioning System, along with traditional

techniques to collect data is providing a new view on

the agricultural process. In this context, this work

provides relevant contributions to the application

domain. The contributions can be described in two

senses:

a) By Ceres System - The application developed is

capable of assisting specialists by providing

facilities to interpret soil analysis and to recommend

the use of fertilizers and limestone, quickly and

reliably. This contribution should be highlighted in

view of the lack of computational solutions in the

application domain, especially in the region where

the system proposed is being used. In this way, the

existence of computational solutions to facilitate the

process is a valuable contribution to the application

domain. Another benefit of the system is the

viability of the use associated to the technologies

used. This allows the development of other

ontologies that will integrate the knowledge base

similarly as presented.

Besides, the knowledge base handling and

instantiated in Protégé through Algernon has

resulted in a reduced number of necessary rules to

inference. This occurs due to the direct use of the

ontology in reasoning execution and by semantic

representativity between the terms defined in the

ontology.

b) By knowledge formalized - The ontology

proposed will allow the development of other

computer applications where the knowledge

modelled also may be used, i.e. the ontology may

provide, as many would hope, the needed

methodology and standard to achieve the objective

of building flexible solutions. Actually, the ontology

has 759 instances of concepts that address of the

interpretation and recommendation processes,

considering the soil’s chemical aspects.

Finally, the work proposed presents benefits in

the computational context by providing modelled,

formalized and validated knowledge contributing to

the development of the new applications in the

agricultural domain. Also, it provides a framework

already validated using technologies combined that

produces good results. On the other hand, to the

agricultural domain, this work contributes by

providing a powerful tool to help the specialist in the

recommendation and interpretation processes,

making them faster and more reliable.

REFERENCES

Roza, D., 2000. Novidade no campo: Geotecnologias

renovam a agricultura. Revista InfoGEO, n 11 -

jan/fev.

Fileto, R., Assad, M. L., Silva, J. V., Soares, A. F.,

Vendrusculo, L.G., 2005. Uma Arquitetura para

Sistema de Informação sobre Solos para o

Zeoneamento Agrícola. Congresso da Sociedade

Brasileira de Informática Agropecuária, Londrina.

Molin, J.P., 2000. Geração e interpretação de mapas de

produtividade para agricultura de precisão. In:.

BORÉM, A. et al. (Ed.). Agricultura de precisão.

Viçosa: Editora UFV. p. 237-57.

Santos, Cristina P., Silva, Denílson R., Silva, Gleidson A.

C., 2007. ONIAQUIS – Uma Ontologia Para

Interpretação de Análise Química do Solo. In: VI

Simpósio de Informática da Região Centro do Rio

Grande do Sul. Unifra. Santa Maria.

Schreiber, G., Akkermans, H., Anjewierden, A., De Hoog,

R., Shadbolt, N., Van de Velde, W., Wielinga, B.,

2000. Knowledge engineering and management: the

CommonKADS methodology. MIT Press.

Gómez-Pérez, A., Fernández-Lópes, M., Corcho, O.,

2003. Ontological Engineering: With Examples from

the Areas of Knowledge Management, e-Commerce

and Semantic Web. Springer-Verlag.

Leão, B. F., A. F. Rocha., 1990. Proposed Methodology

for knowledge aquisition: a study on congenital heart

disease diagnosis. Methods of Information in

Medicine.

Noy, N.F., Crubezy, M., Fergerson, R.W. et al. (2003)

Protégé-2000: an open-source ontology-development

INTERPRETATION AND RECOMMENDATION TASKS SUPPORTED BY CERES SYSTEM

467