A Cognition-inspired Knowledge Representation Approach for

Knowledge-based Interpretation Systems

Joel Luis Carbonera and Mara Abel

Institute of Informatics, Federal University of Rio Grande do Sul, Porto Alegre, Brazil

Keywords:

Knowledge Representation, Reasoning, Ontology, Petroleum Geology.

Abstract:

We propose a hybrid approach for knowledge representation that combines classic representations (such as

rules and ontologies) with cognitively plausible representations, such as prototypes and exemplars. The re-

sulting framework can be used for developing knowledge-based systems that combine knowledge-driven and

data-driven techniques. We also present how this approach can be used for developing an application for

interpretation of depositional processes in Petroleum Geology.

1 INTRODUCTION

Nowadays, knowledge-based systems can be viewed

as a mature technology, which have been applied

for supporting the decision making in a wide range

of scenarios. In general, knowledge-based systems

are characterized by a top-down approach, where

the relevant knowledge is explicitly represented in a

computer-processable way and the problem-solving

process performed by the system is supported by this

knowledge. Recently, ontologies have been applied

for representing the domain knowledge in knowledge-

based systems. Since an ontology speciﬁes in a for-

mal and explicit way the shared conceptualization in

a given domain (Studer et al., 1998), its adoption al-

lows the knowledge reusing and promotes the seman-

tic interoperability among systems.

Knowledge-based systems can be applied when

there is no data available for applying bottom-up

approaches, such as those based on machine learn-

ing techniques. Besides that, in general, the results

achieved by knowledge-based systems are reliable,

since they are obtained by applying knowledge in

which the practitioners of the domain rely on. How-

ever, for developing these systems it is necessary to

elicit the relevant knowledge from reliable knowledge

sources (domain experts, domain literature, etc). This

is one of the weak points of adopting this approach,

since the knowledge acquisition is a notoriously difﬁ-

cult task, involving costly and error-prone processes

that, in general, depends on the availability of do-

main experts. Besides that, sometimes, the elicited

knowledge is not enough to cover all the situations

with which the application should deal.

In this position paper, we claim that these weak-

nesses of the knowledge-based systems can be mit-

igated by providing to the system the capability of

extracting useful knowledge from the available data;

that is, from previously solved instances available to

the system. The resulting approach can be viewed

as a combination of knowledge-driven (top-down)

and data-driven (bottom-up) approaches. Hybrid ap-

proaches like this were already proposed in the liter-

ature (Li and Love, 1999). However, in this paper,

we go a step further, by proposing a hybrid cognition-

inspired knowledge representation approach for sup-

porting this kind of system.

Within the cognitive sciences there is an ongoing

debate concerning how the knowledge is represented

in the human mind. According to (Murphy, 2002),

there are three main theories in this debate. The clas-

sical theory assumes that each concept is represented

by a set of features that are shared by all the entities

that are abstracted by the concept. In this way, this set

of features can be viewed as the necessary and suf-

ﬁcient conditions for a given entity to be considered

an instance of a given concept. In this theory, con-

cepts are viewed as rules for classifying objects based

on features. The prototype theory, on the other hand,

states that concepts are represented through a typi-

cal instance, which has the typical features of the in-

stances of the represented concept. Finally, the exem-

plar theory assumes that each concept is represented

by a set of exemplars of it, which are explicitly rep-

resented in the memory. These exemplars are enti-

ties that were previously experienced by the agent. In

644

Carbonera J. and Abel M..

A Cognition-inspired Knowledge Representation Approach for Knowledge-based Interpretation Systems.

DOI: 10.5220/0005467106440649

In Proceedings of the 17th International Conference on Enterprise Information Systems (ICEIS-2015), pages 644-649

ISBN: 978-989-758-096-3

 2015 SCITEPRESS (Science and Technology Publications, Lda.)

theories based on prototypes or exemplars, the cat-

egorization of a given entity is performed according

to its similarity with prototypes or exemplars; the in-

stance is categorized by the category that has a pro-

totype (or exemplar) that is more similar to it. There

are some works, such as (Fiorini et al., 2014), which

apply these alternative theories in computer applica-

tions.

Our main proposal in this position paper is to

combine classic representations (such as ontologies

and rules), prototypes and exemplars in a knowledge

representation framework for supporting knowledge-

based systems. The resulting framework combines

the strengths of the knowledge-driven (top-down) and

data-driven (bottom-up) approaches, overcoming the

weaknesses of both. In our framework, part of the

knowledge is represented classically, using ontologies

and rules, and part of the knowledge is represented

as prototypes and exemplars, which can be extracted

from the data that is processed by the system. Thus,

our approach can provide the reliance of the expert

knowledge when it can be applied, however, it also

is able to provide solutions for cases that cannot be

covered by this knowledge, but that can be estimated

from the available data. In this approach, the result-

ing system can perform problem-solving processes

by combining rule-based reasoning with similarity-

based reasoning. The rule-based reasoning can be

used ﬁrst for reducing the search space and, if a suit-

able solution is not found, a similarity-based reason-

ing component can be triggered for ﬁnding the suit-

able solution, by comparing the instance that is being

analyzed with the available prototypes and exemplars.

In our project, our main focus of interest are

knowledge-based systems for data interpretation, in

the domain of Petroleum Geology. In this paper,

we discuss how to use our approach for developing

a knowledge-based system for visual interpretation

of depositional processes. This task is a resource-

consuming job, which relies intensively on the visual

knowledge of geologists, and that is considered a cru-

cial step in petroleum exploration.

2 SEDIMENTARY

STRATIGRAPHY

Since we will discuss an application for the Sedimen-

tary Stratigraphy ﬁeld, in this Section, we will present

an overview of the domain. Sedimentary Stratigraphy

is a sub-ﬁeld of Geology that studies the sedimentary

terrains in the surface or subsurface of the Earth, in

order to determine the geological history of their for-

mation. The main objects of study and description

are Body of Rock, Well Core, Outcrop, Sedimentary

Facies, Sedimentary Structures and Depositional Pro-

cesses. A body of rock can be a well core, which is

a cylinder of rock extracted from the subsurface by

means of drilling; or an outcrop, which is a body of

rock exposed in the surface. A sedimentary facies is a

region in a body of rock, visually distinguishable from

adjacent regions. Each sedimentary facies is a di-

rect product of a depositional process. A sedimentary

structure is the external visual aspect of some internal

spatial arrangement of the rock grains. Finally, de-

positional processes are events that involve the com-

plex interaction of natural forces and sediments, and

which are responsible for the formation of sedimen-

tary rocks. Figure 1 presents an example of a well

core, emphasizing two distinct sedimentary facies.

Figure 1: a well core, emphasizing two distinct sedimentary

facies (adapted from (Lorenzatti et al., 2009)).

In the task of visual interpretation of depositional

processes, the geologists visually inspect sedimentary

facies (in well cores or outcrops) and interpret which

was the corresponding depositional process that was

responsible by the formation of this facies.

3 OUR APPROACH

In this section, we will present our approach and how

it can be applied in a knowledge-based system for vi-

sual interpretation of depositional processes.

In our approach, we assume the availability of

static knowledge about the domain, represented as a

domain ontology. In our application, we adopted a do-

main ontology for Sedimentary Stratigraphy. Figure

2 represents an excerpt of our domain ontology, pre-

senting the core domain entities that were explained

in Section 2. Figure 3 represents an excerpt of the

taxonomy of depositional processes.

This domain ontology is used in our application

for providing a formal basis for describing facts about

the domain and for articulating the inferential knowl-

edge (rules), which are used for performing the in-

terpretation. Our system takes as inputs descriptions

ACognition-inspiredKnowledgeRepresentationApproachforKnowledge-basedInterpretationSystems

645

Chemical and diagenetic structure

Depositional structure

Sedimentary faciesBody of rock

Sedimentary structure

Biogenic structure

Deformation structure

Well core Depositional processOutcrop

hasSedimentaryStructure

generatedBy

hasSedimentaryFacies

Figure 2: Representation of the core domain entities of the

ontology. This excerpt omits the attributes of the entities

and the other sub-types of sedimentary structure.

of bodies of rock, which are made according the on-

tology. When the user requires the execution of the

interpretation process, the application analyses each

sedimentary facies of the body of rock, interpreting

the respective depositional process that have gener-

ated it.

Regarding the visual interpretation task performed

by our system, following (Carbonera et al., 2015; Car-

bonera et al., 2013; Carbonera et al., 2011), we also

consider that it is inferentially characterized by the

application of rules in the form of

observation ⇒ interpretation (1)

, where observation is a set of statements that de-

scribe the features that should be perceived in some

observable entities for supporting the interpretation,

and interpretation is a set of statements about inter-

pretable entities. In our application, the observable

entities are sedimentary facies and sedimentary struc-

tures and the interpretable entities are depositional

processes. In this way, the observation comprises

statements about features that should be perceived in

a given instance f of sedimentary facies for inferring

the interpretation, which speciﬁes what is the spe-

ciﬁc type of depositional process (see Figure 3) that

was responsible by generating f . Following, it is pre-

sented an example of rule used in our application:

∀ f , s, p SedimentaryFacies( f ) ∧ hasSorting( f , PoorlySorted) ∧

hasLithology( f , Sandstone) ∧ TroughCrossStratiﬁcation(s) ∧

hasSedimentaryStructure(f , s) ∧ DepositionalProcess(p) ∧

generatedBy( f , p) → 3dDuneMigration(p)

(2)

This rule represents that a sedimentary facies f , which

is poorly sorted, whose lithology is sandstone and that

has a sedimentary structure s, which is a Trough-cross

Stratiﬁcation (a sub-type of sedimentary structure);

was generated by a depositional process p, which is

a 3D dune migration (a sub-type of depositional pro-

cess).

It is important to notice that, in our domain of

interest, it is necessary to provide to the system the

capability of providing interpretations in several lev-

els of generalization/speciﬁcity. This is important be-

cause sometimes there is not enough information in

Figure 3: Representation of an excerpt of the taxonomy of

depositional processes.

the sedimentary facies description for supporting the

interpretation of the speciﬁc type of depositional pro-

cess that have generated the facies. We provide this

capability to our application by including a taxonomy

(with several levels of generalization) of depositional

processes in the domain ontology and by including

the necessary rules for interpreting each different type

of depositional process speciﬁed by this taxonomy.

The ontology, the rule base and the rule-based

reasoning engine constitute the knowledge-driven

component of our system. In general, it can be very

effective in reducing the set of alternative possible in-

terpretations of a given facies. It can prune entire

branches of alternatives in the taxonomy of deposi-

tional processes. It can even reach speciﬁc interpreta-

tions (some leaf of the taxonomy, such as 3D dune

migration), if the description of sedimentary facies

is detailed enough. However, this is not always the

case. It is common the absence of important infor-

mation in facies descriptions, because some features

are not identiﬁable by the geologist in the object, or

because some identiﬁed features are omitted during

the description. Besides that, the inferential knowl-

edge can be incomplete, in a way that the set of rules

cannot cover all the possible cases that arise in real

scenarios. In these cases, the knowledge-driven com-

ponent can only provide a more general interpretation

(such as Subaqueous tractive current, for example).

For overcoming these weaknesses, we propose a

data-driven component that acts in conjunction with

the knowledge-driven component. This component

relies on a knowledge representation approach that is

inspired by the prototype theory and exemplar theory

of knowledge representation in the human mind. Fol-

lowing this idea, we assume that for each possible rel-

evant interpretable entity ie, the system has a proto-

type and a set of exemplars of the observations that

can support the interpretation of ie. In this compo-

nent, the visual interpretation is carried out through

similarity-based reasoning, where the instances that

should be interpreted are compared with the proto-

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

646

types and exemplars for determining the suitable in-

terpretation. In addition, since the system should ex-

tract the prototypes and exemplars from the available

data (that can include instances that were previously

solved by the system); it includes a component for ex-

tracting the prototypes and exemplars: the knowledge

extractor.

Before presenting the details of the data-driven

component, let us consider a formal characterization

of some important notions.

• DS = {r

, r

, ..., r

} is a data set, which

can be abstractly considered as a set of

n records. Each r

can be viewed as

a 2-tuple (observation, inter pretation); where

interpretations are articulated using concepts in

the domain ontology that are generically called

Interpretable entities, and observations are artic-

ulated using concepts in the domain ontology that

are generically called observable entities.

• O

= {o

, o

, ..., o

} is the set of the

observations of the records in DS, in a way

that o

∈ O

is the observation of the record

∈ DS. There ir one observation for each record.

• ro: DS → O

is a function that maps a given

record r

∈ DS to its observation o

∈ O

• IE = {ie

, ie

, ...ie

} is a set of m interpretable

entities, where each ie

is a concept provided by

the domain ontology. The elements of IE are or-

ganized in a taxonomy, which is a partially or-

dered set (IE, ≤), where ≤ is the subsumption re-

lation held between elements of IE.

• P = {p

, p

, ..., p

} is a set of m prototypes of

observations, where each p

represents the typi-

cal observation whose interpretation is speciﬁed

by the interpretable entity ie

• E = {es

, es

, ..., es

} is a set whose elements

are sets of exemplars, such that, each es

, e

, ..., e

} is a set of k exemplars of observa-

tions whose interpretation is speciﬁed by the in-

terpretable entity ie

• prototype : IE → P is a function that maps a given

interpretable entity ie

∈ IE to the prototype p

∈

P, which represents the typical observation hose

interpretation is speciﬁed by ie

∈ IE.

• exemplars: IE → E is a function that maps a

given interpretable entity ie

∈ IE to the set es

of exemplars, such that es

∈ E, which represents

the set of k exemplars of observations whose in-

terpretation is speciﬁed by the interpretable entity

• records: IE → 2

is a function that maps a given

interpretable entity ie

∈ IE to the set DS

⊆ DS

of records in DS, which represents the set of

records in DS whose interpretation is speciﬁed by

the interpretable entity ie

• obs : IE → 2

is a function that maps a given

interpretable entity ie

∈ IE to a set O

⊆ O

which represents the set of observations related to

the records speciﬁed by records(ei), where ∀o

∈

, ∃r

∈ records(ei), o

= ro(r

The user data in the system database is stored as

a set of instances of the domain ontology. However,

for the purposes of the data-driven component, this

database can be abstractly viewed as the dataset DS,

where observations (including individual exemplars

of observations in the set E, and typical observations

in the set P) can be viewed as vectors of features of

length l, called observation vectors. The vectorial

representation is preferable for understanding how to

perform similarity judgments. Figure 4 presents the

process of representing instances of our domain ontol-

ogy as vectors of features. Notice that, in our applica-

tion, observations consist basically in descriptions of

instances of sedimentary facies (our observable enti-

ties), and interpretations are basically types of depo-

sitional processes (our interpretable entities). Thus,

in the resulting observation vectors, each attribute of

Sedimentary Facies is represented as a position of the

resulting vector of features. Besides that, the result-

ing vector has a special position for representing the

type of the Sedimentary Structure of the Sedimentary

Facies.

Figure 4: Representation of the process of converting the

instance of Sedimentary facies (represented according the

domain ontology) in a vector of features.

The extraction of prototypes of observations from

data is performed by the knowledge extractor (in the

knowledge-driven component). This component ex-

tracts a prototype of the observations whose interpre-

tation is some interpretable entity ie (obs(ie)). It ba-

sically analyses the observations in obs(ie) and deter-

mines the typical value of each one of the l attributes

of the observations. If the attribute is numeric, the

typical value can be the average; if the attribute is

nominal, the typical value can be the most frequent.

The extraction of a set of exemplars of obser-

ACognition-inspiredKnowledgeRepresentationApproachforKnowledge-basedInterpretationSystems

647

vations from data is also performed by the knowl-

edge extractor, in the knowledge-driven compo-

nent. Considering a given ie ∈ IE, the exemplars

in exemplars(ie) will be used by the system as ref-

erences for interpreting a new observation o

new

, by

performing judgments of the similarity between o

new

and each considered exemplar in exemplars(ie). In

our approach, we assume that it is not desirable to

consider all records in obs(ei) as exemplars for rep-

resenting the observations in obs(ei), since the com-

putational cost of the interpretation process is pro-

portional to the number of exemplars that are se-

lected for representing the observations. For reduc-

ing the computational overhead of the interpretation

process, in our approach we consider that the num-

ber of exemplars related to each interpretable en-

tity ei

∈ IE is deﬁned as a percentage ep (deﬁned

by the user) of the total number of observations in

obs(ei

). This raises the problem of how to select

which observations in obs(ei

) will be consider as

the exemplars in exemplars(ei

). We select three

main criteria that an observation o

∈ obs(ei

) should

meet for being included in exemplars(ei

), consider-

ing an interpretable entity ie: (i) o

should have a

high degree of dissimilarity with the prototype given

by prototype(ie); (ii) o

should have a high degree

of similarity with a big number of observations in

obs(ei

); and (iii) o

should have a high degree of

dissimilarity with each exemplar already included in

exemplars(ei

). This set of criteria was developed for

ensuring that the set of exemplars in exemplars(ei

)

will cover in a reasonable way the spectrum of vari-

ability of the observations in obs(ei

). That is, our

goal is to preserve in exemplars(ei

) some uncom-

mon observations, which are not well represented by

prototype(ie), but that represent the variability of the

observations. In our approach, we apply these crite-

ria, by including in exemplars(ei

) the h ﬁrst observa-

tions from obs(ei

) that maximize their exemplariness

index. The exemplariness index is computed using the

notion of density of a given observation (represented

in the form of an observation vector). Regarding some

interpretable entity ie, the density of some observa-

tion o

∈ obs(ie), is computed by the function density,

such that, density(o

) = −

|obs(ie)|

∑

|obs(ie)|

j=1

d(o

, o

);

where d is some dissimilarity (or distance) function (a

function that measures the dissimilarity between two

entities). Considering this, the set exemplars(ei

) of

some interpretable entity ie, with h exemplars, can be

computed by the Algorithm 1.

Notice that the Algorithm 1 basically selects from

obs(ie), the observations that maximize the exemplar-

iness index, which is the sum of: (i) distance (or dis-

similarity) of the observation from the prototype(ie);

Algorithm 1: extractExemplars.

Input: An interpretable entity ie and a number h of exemplars

Output: A set exemplars

of h observations representing the

exemplars of the observations whose interpretation is

speciﬁed by ie.

begin

exemplars

← ∅;

for j ← 1 to h do

eIndex

max

← −∞;

obs

max

← null;

foreach o

∈ obs(ie) do

density ← density(o

);

d p ← d(o

, prototype(ie));

med ← 0;

if exemplars

is not empty then

Compute the distance between o

and each exemplar e

already included in exemplars

and assign to med the

distance of the nearest exemplar from o

;

/* eIndex is the exemplariness index */

eIndex = d p + density + med;

if eIndex > eIndex

max

then

eIndex

max

← eIndex;

obs

max

← o

;

exemplars

← exemplars

∪ obs

max

;

return exemplars

;

(ii) the density of the observation, considering the set

obs(ie); and the distance (or dissimilarity) of the ob-

servation from its nearest exemplar, already included

in exemplars

Once the system has the hybrid knowledge rep-

resentations (prototypes and exemplars), they can be

applied by the interpretation engine for performing

interpretations. This component takes as input a

new observation and provide its corresponding inter-

pretation. In our application, the observation is an

ontology-based description of a sedimentary facies,

and the interpretation is a speciﬁc type of depositional

process that was responsible by the formation of the

facies. Firstly, the interpretation engine applies rule-

based reasoning (using classical knowledge represen-

tations in the form of a domain ontology and a rule

base) for providing a ﬁrst set of hypothesis. Notice

that the rule-based reasoning can infer more than one

interpretation for the same observation, depending on

the rules in the rule base. If the rule-based reason-

ing provides interpretations that are not speciﬁc (the

leaves of the taxonomy of interpretable entities in the

domain ontology), the similarity-based reasoning can

be used for determining more speciﬁc interpretations.

The visual interpretation engine implements the Al-

gorithm 2.

Notice that the Algorithm 2 uses the notion of ap-

plicability, which, intuitively measures the degree in

that a given interpretable entity ie can be applied as

an interpretation for a given observation obs. The ap-

plicability is computed by the Algorithm 3, using the

ICEIS2015-17thInternationalConferenceonEnterpriseInformationSystems

648

Algorithm 2: visualInterpretation.

Input: An observation obs.

Output: A set int

set

of interpretable entities representing the

interpretations of obs.

begin

int

set

← ∅;

Perform the rule-based reasoning for interpreting obs (applying

the rules in the rule base), and include the interpretable entities of

the resulting interpretations in int

set

;

if the interpretable entities in int

set

are not speciﬁc then

hyp

set

← ∅;

foreach ie ∈ int

set

Find the leaves in the taxonomy of interpretable entities,

whose root is ie, and include them in hyp

set

;

int

set

← ∅;

MAX ← −∞;

foreach ie ∈ hyp

set

app ← applicability(ie, obs);

if app > MAX then

MAX ← app;

int

set

← ∅;

int

set

← int

set

∪ ie;

else if app = MAX then

int

set

← int

set

∪ ie;

return int

set

;

Algorithm 3: applicability.

Input: An interpretable entity ie and an observation obs.

Output: A value r ∈ R, which is the degree in that ie can be applied

as an interpretation for obs.

begin

app ← 0;

pSimilarity ← sim(obs, prototype(ie));

eSimilarity ← 0;

Calculate the similarity sim(obs, ex

) between obs and each

∈ exemplars(ie), and assign to eSimilarity the similarity value

of the most similar ex

;

app ← pSimilarity + eSimilarity;

return app;

prototypes and exemplars of the observations in

obs(ie).

Notice that the Algorithm 3 uses the function sim

for measuring the similarity. Intuitively, the similar-

ity is the inverse of the dissimilarity (or distance) be-

tween two observations. Thus, sim has values that

are inversely proportional to the values obtained by

the function d. Here, we assume that sim(o

, o

) =

exp(−d(o

, o

)).

4 CONCLUSIONS AND FUTURE

WORKS

In this paper, we propose a cognition-inspired

knowledge representation framework for supporting

knowledge-based systems. The resulting framework

can combine the strengths of the knowledge-driven

(top-down) approaches, overcoming the weaknesses

of both. In future works, we intend to provide a de-

tailed account of the whole systems developed us-

ing this approach, and to present a validation of the

approach in real cases. In addition, we plan to in-

vestigate how the system can update its representa-

tions (prototypes and exemplars), when new data is

included in its database.

ACKNOWLEDGEMENTS

We gratefully thank to PRH PB-217 program (sup-

ported by Petrobras), for the ﬁnancial support to this

work, and to Sandro Fiorini for comments and ideas.

REFERENCES

Carbonera, J. L., Abel, M., and Scherer, C. M. (2015). Vi-

sual interpretation of events in petroleum exploration:

An approach supported by well-founded ontologies.

Expert Systems with Applications, 42:2749 – 2763.

Carbonera, J. L., Abel, M., Scherer, C. M., and Bernardes,

A. K. (2013). Visual interpretation of events in

petroleum geology. In Proceedings of ICTAI 2013.

Carbonera, J. L., Abel, M., Scherer, C. M. S., and

Bernardes, A. K. (2011). Reasoning over visual

knowledge. In Vieira, R., Guizzardi, G., and Fior-

ini, S. R., editors, Proceedings of Joint IV Seminar on

Ontology Research in Brazil, volume 776.

Fiorini, S. R., Abel, M., and Carbonera, J. L. (2014). Rep-

resentation of part-whole similarity in geology. Earth

Science Informatics, Special Issue on Semantic e-

Sciences.

Li, H. and Love, P. E. (1999). Combining rule-based expert

systems and artiﬁcial neural networks for mark-up es-

timation. Construction Management & Economics,

17(2):169–176.

Lorenzatti, A., Abel, M., Nunes, B. R., and Scherer, C.

M. S. (2009). Ontology for imagistic domains: Com-

bining textual and pictorial primitives. In ER Work-

shops.

Murphy, G. L. (2002). The big book of concepts. MIT press.

Studer, R., Benjamins, V. R., and Fensel, D. (1998). Knowl-

edge engineering: Principles and methods. Data and

Knowledge Engineering, 25(1-2):161–197.

ACognition-inspiredKnowledgeRepresentationApproachforKnowledge-basedInterpretationSystems

649