OACAS

Ontologies Alignment using Composition and Aggregation of Similarities

Sami Zghal

CRIL CNRS FRE 2499, Artois University, IUT of Lens, Rue de l’Universit

e - S.P. 16, 62307 Lens Cedex, France

Marouen Kachroudi, Sadok Ben Yahia

Computer Science Department, Faculty of Sciences of Tunis, Campus Universitaire, 1060 Tunis, Tunisia

Engelbert Mephu Nguifo

LIMOS CNRS UMR 6158, Complexe scientiﬁque des C

ezeaux, BP 125, 63173 Aubiere Cedex, France

Keywords:

Ontology alignment, Similarity measures, Composition of similarities.

Abstract:

Ontologies are the kernel of semantic Web. They allow the explicitation of the semantic purpose for structur-

ing different ﬁelds of interest. In order to harmonize them and to guarantee the interoperability between these

resources, the topic of alignment of ontologies has emerged as an important process to reduce their heterogene-

ity and improve their exploitation. The paper introduces a new method of alignment of OWL-DL ontologies,

using a combination and aggregation of similarity measures. Both ontologies are transformed into a graph

which describes their information. The proposed method operates in two steps: local (linguistic similarity

composition and neighborhood similarity) step and the aggregation one.

1 INTRODUCTION

An ontology is deﬁned as ”an explicit speciﬁcation of

a conceptualization” (Gruber, 1993). Indeed, an on-

tology is a set of concepts, relations and possibly ax-

ioms representing a knowledge ﬁeld. Consequently,

the diversity of the reality is at the same time a source

of richness of and heterogeneity. This heterogeneity

unfortunately reduces the interoperability levels (Eu-

zenat, 2001). Thus, the process of ontologies align-

ment aims to lower the conﬂict between them.

The ontology alignment issues grasped the interest

of the community as witnesses the wealthy number

of approaches, e.g., FALCON-AO (Hu et al., 2007),

ONTODNA (Kiu and Lee, 2007) and RIMOM (Li

et al., 2007), to cite but a few. The FALCON-AO sys-

tem, contains ﬁve modules integrating a graphic in-

terface. The process begins with a ”parsing” stage

to extract a graph model representing the character-

istics of the ontology. The following stage consists

in choosing the strategy of alignment, through a li-

brary of aligners. Their role is the exploitation of the

properties of the ontology. The resultant alignment

is presented under the format RDF / XML, having

considered the linguistic aspect as well as the struc-

tural aspect of an ontology. The second system, ON-

TODNA uses techniques of data mining to tackle the

issue of semantic heterogeneity. This system operates

in four stages. The algorithm takes in entry two on-

tologies to be aligned. Then, it launches a process of

linguistic and lexical treatments. A stage of Cluster-

ing is performed according to the best obtained corre-

spondences. A ﬁnal linguistic treatment is applied in

the sake of restoring the semantic relations between

the various ontological entities. The RIMOM sys-

tem is an alignement tool which contains six strate-

gies. Every strategy is deﬁned according to the type

of information that an ontology can contain. Indeed,

the system offers seven different alignment methods,

which are afterward organized through a linear inter-

polation function. The system also exploits the struc-

tural aspect of an ontology, by the propagation of

similarity through its hierarchy. The ﬁnal alignment

is obtained after a sequence of reﬁnements, through

heuristic rules to keep the best possible alignment.

The new alignment method, OACAS (Ontolo-

gies Alignment using Composition and Aggregation

of Similarities), introduces a new alignment algorithm

of OWL-DL (Ontology Web Language Description

Logic) ontologies. The main thrust of this method is

233

Zghal S., Kachroudi M., Yahia S. and Nguifo E. (2009).

OACAS - Ontologies Alignment using Composition and Aggregation of Similarities.

In Proceedings of the International Conference on Knowledge Engineering and Ontology Development, pages 233-238

DOI: 10.5220/0002301802330238

 SciTePress

the application of the most suitable similarity measure

depending of the category of the node in the ontol-

ogy. In addition, the OACAS method explores a wider

neighborhood than do the pioneering methods of the

literature. Carried out experiments showed that OA-

CAS presents very encouraging values of the com-

monly used evaluation metrics for the assessment of

ontologies alignement, especially for the real ontolo-

gies.

The paper is organized as follows. Section 2 intro-

duces the new ontology alignment method and thor-

oughly presents the underlying fundamentals. Sec-

tion 3 presents an evaluation of the new method that

includes a description of the evaluation metrics used,

experimental tests and a comparative study. The con-

clusion and future issues are presented in section 4.

2 THE OACAS METHOD

The proposed new ontology alignment method, OA-

CAS, aligns two ontologies. Both ontologies are de-

scribed in the OWL-DL language. Both ontologies

are transformed in two graphs O-GRAPHS. The ob-

tained graphs are parsed in order to produce the align-

ment process out. The process of building the graphs

allows to map the considered ontologies to be aligned

in two graphs, called O-GRAPHS. An O-GRAPH

describes all the information included in an OWL-

DL ontology: classes, relations and instances. Both

classes and instances represent the nodes of the graph.

The relations between these different entities are in-

duced by the links of an O-GRAPH. Each entity of

the ontology is formalized through an associated no-

tion to the RDF formalism. OWL-DL ontology enti-

ties are described thanks to OWL language construc-

tors. These constructors are represented through RDF

triplets: <subject, predicate, object>. In an OWL-

DL ontology, a class or a relation description is an

RDF triplet. The subject corresponds to the class or

to the relation. Predicates are OWL primitives, which

are OWL and RDF properties. Each property, used in

a triplet, sketches a knowledge of the described entity.

The arrangement of tho se knowledge constitues the

entity deﬁnition. The representation of an OWL-DL

ontology through an O-GRAPH permits to load the

ontology in main memory only once. An O-GRAPH,

stored in main memory, statistically reduces the time

required to access initial OWL-DL ontology disk res-

ident ﬁle.

2.1 The Alignment Method

The introduced OACAS method lays on a compo-

sition of similarity computation based model. The

method starts by exploring the O-GRAPH structure. It

determines the nodes of both ontologies to be aligned

and gets out the similarity measures. For each node of

the same category (or cluster), the alignment model

computes similarity mesures between descriptors by

using appropriate functions. An aggregation func-

tion combines the similarity measures and the node’s

structures of the nodes to be aligned. Thus, this

function considers all the descriptive information of

this couple (name, comment and label) as well as its

neighborhood structure. The algorithm implementing

the OACAS method takes as input two OWL-DL on-

tologies to be aligned and produces an RDF ﬁle con-

taining the aligned nodes as well as their similarity

measures. The OACAS method operates into two suc-

cessive steps. The ﬁrst one computes the local simi-

larity, whereas the second one computes the aggrega-

tion similarity.

2.1.1 The Local Similarity Computation

The local similarity computation is performed into

two successive steps. The ﬁrst step computes many

linguistic similarity measures and aggregates them for

each couple of nodes belonging to the same category

(or type). The second step computes neighborhood

similarities by exploiting the structures of the nodes

to be aligned.

The Linguistic Similarity Composition. The lin-

guistic similarity computation is carried out once for

each node of the same cluster (node of the same type)

in the beginning of the alignment process. The lin-

guistic similarity measures of couples of entities of

the same type (class, property and instance) are com-

puted using the LINGUISTIC function. The names of

properties and instances are used to compute linguis-

tic similarities. For class category, the computation of

the linguistic similarity considers both the comments

and labels. The computation of linguistic similarities

uses different similarity measures. Those measures

are adapted to different descriptors (names, comments

and labels) of the entities to be aligned. Different sim-

ilarity values obtained, for the descriptors, are com-

posed. This composition assigns weights to each sim-

ilarity measure of descriptors. The sum of the as-

signed weights to different similarity values is equal

to 1. This unit sum guarantees that the composition of

the similarity produces a normalized value (between

0 and 1). The LEVENSHTEIN similarity measure (Eu-

zenat and Shvaiko, 2007) is used to compute the simi-

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

234

larity value between the names of ontological entities.

The Q-GRAM similarity measure (Ukkonen, 1992)

computes the similarity value between the comments

of the ontological entities. The JARO-WINKLER simi-

larity measure (Euzenat and Shvaiko, 2007) computes

the similarity value between the labels of ontological

entities. The LINGUISTIC function computes com-

posed linguistic similarity of couples of nodes of both

ontologies to be aligned, i.e., O

and O

. It takes as in-

put (i) both ontologies sketched by two corresponding

O-GRAPHS; (ii) linguistics similarity functions (i.e.,

Funct); and (iii) weighted attributed to the descriptors

nodes (i.e., Π

). As a result, it produces a composed

linguistic similarity vector, V

CLS

, for each couple of

nodes. The similarity function Funct considers two

nodes, N

and N

, and returns the linguistic similarity

value of the descriptor, Sim

. LEVENSHTEIN or Q-

GRAM or JARO-WINKLER implements the similarity

function, Funct, depending of the type of the nodes.

Composed linguistic similarity, Sim

, is computed

depending of the descriptors of nodes to be aligned

and associate weights to each descriptor, Π

. Both

nodes (N

and N

) and the associated composed lin-

guistic similarity (Sim

) are added to the composed

linguistic similarity vector (V

CLS

). The composed lin-

guistic similarity of different couples of entities will

be used to compute the neighborhood similarity as

sketched in the following.

The Neighborhood Similarity Computation. The

NEIGHBORHOOD function considers both ontologies

to be aligned (i.e., O

and O

), the composed similar-

ity vector (V

CLS

), the weights assigned to each cate-

gory (Π

) and the weights associated to the neighbor

level (Π

). Therefore, it produces the neighborhood

similarity vector, V

. The neighborhood similarity

computation needs composed linguistic similarity of

the couple of nodes to be aligned and the nodes struc-

tures. Neighborhood nodes are organized by category,

node having the same type. The neighborhood simi-

larity computation propagates similarity into two suc-

cessive neighborhood levels. The ﬁrst level (level 1)

includes direct neighbors of the nodes to be aligned

whereas second one (level 2) contains indirect neigh-

bors. Direct neighbors of the ﬁrst level represent

nodes having direct relationship with the node under

consideration. Neighbors of the second level repre-

sent nodes having relationship with the nodes of the

ﬁrst one. The neighbors entities of the ﬁrst level are

clustered into three categories (classes, instances or

properties). Each category (or cluster) includes onto-

logical entities having the same type. After the step of

clustering, the neighborhood similarity is computed

between those categories. The neighborhood nodes

of the level 2 are treated in the same manner as the

neighbors of the ﬁrst one. The neighborhood similar-

ity by group MSim takes nodes from vectors V N

and

V N

regrouped by category (where VN

and V N

de-

note a vector nodes of O

and O

). The process com-

putation uses the ”Match-Based similarity” (Valtchev,

1999) as follows:

MSim(E, E

) =

∑

(i,i

)∈Pairs(E,E

)

Sim

CLS

(i, i

)

Max(|E|, |E

. (1)

Both sets E and E

represent nodes of the same

cluster belonging respectively to vectors V N

and

V N

. The neighborhood similarity, Sim

, is com-

puted using Equation 2:

Sim

∑

i∈(1,2)

(Π

(

∑

(E,E

)

(E,E

)

MSim(E, E

))), (2)

where i stands for the level (i.e., 1 or 2). The

neighborhood similarity, Sim

is a normalized value,

since the sum of weights assigned to different neigh-

bors is equal to 1, (Π

V 1

+ Π

V 2

= 1). Direct neigh-

bors (level 1) have more important relationships than

those of indirect one (level 2). Thus, nodes of level 1

have an important impact on the produced alignment.

For this reason, the weight assigned to the ﬁrst level,

V 1

= 0.8, is more important than the one assigned to

the second level, Π

V 2

= 0.2. In addition, the sum of

weights assigned to the category of nodes is equal to 1

(

∑

(Π

) = 1). Those weights are uniformly assigned

between the different categories. The neighborhood

similarity is computed thanks to an iterative process,

level by level. The obtained values of the composed

linguistic similarity, i.e. V

CLS

, and neighbors similar-

ity, i.e. V

, are combined in order to compute aggre-

gation similarity.

2.1.2 The Aggregated Similarity Computation

The aggregation similarity is a combined similarity

between the local similarities (the composed linguis-

tic similarity and the neighborhood similarity). Func-

tion AGGREGATION needs to have in input both on-

tologies to be aligned, O

and O

, the two similar-

ity vectors, V

CLS

and V

, and the weights attributed

to the both kind of similarities, Π

and Π

. It pro-

duces the aggregated similarity vector, V

. For each

couple of entities, N

and N

, of the same category

of the both ontologies to be aligned, O

and O

, the

aggregated similarity is computed as follows:

Sim

, e

) = Π

Sim

, e

) +Π

Sim

, e

). (3)

Note that the sum of the weights, attributed to each

kind of similarity, is equal to 1 in order to have a nor-

malized aggregation (between 0 and 1). In addition,

OACAS - Ontologies Alignment using Composition and Aggregation of Similarities

235

the sum of weights is equal to 1 (Π

+ Π

= 1). In

the next section, we focus on the experimental evalu-

ation of OACAS.

3 EXPERIMENTAL EVALUATION

The carried out experimental evaluation uses the tests

provided and distributed by the OAEI

in order to pro-

mote activities within the ontology alignment com-

munity. The goal of these benchmark bases is to iden-

tify the strength and weakness areas in each alignment

algorithm. The battery of tests is based on one par-

ticular ontology. This ontology is dedicated to the

very narrow domain of bibliography. From this on-

tology (i.e., Test 101) derives a number of alternative

ontologies of the same domain for which alignments

are provided. The benchmark test library is composed

of a set of 51 pairs of ontologies. Each ontology is

to be aligned with the reference ontology (i.e., Test

101). The Test 101 contains 33 named classes, 24 ob-

ject properties, 40 data properties, 56 named individ-

uals and 20 anonymous individuals. The alignment

method should supply for each test an alignment. The

obtained alignment is compared to the available ref-

erence one. Both, obtained and reference alignments,

are used to compute evaluation metrics in order to as-

sess the quality of the aligner algorithm.

3.1 Evaluation Metrics

Precision and recall are respectively the most used

metrics to evaluate the quality of an alignment method

(Euzenat et al., 2006). The OAEI uses these measures

to assess the quality of the obtained alignment. The

main goal of these measures is to assess the autom-

atization of the comparing process of the alignment

methods. The ﬁrst step, within the process of evalua-

tion of the quality of alignment, consists in resolving

the problem manually. The manually obtained result

is considered as the reference alignment. The com-

parison between the reference alignment and that ob-

tained by the alignment method produces three sets:

f ound

, N

expected

and N

correct

. The N

f ound

set repre-

sents the set of the couples aligned by an alignment

method. The N

expected

set gathers the couples aligned

in the reference alignment. The N

correct

set is the in-

tersection of both N

f ound

and N

expected

sets. It repre-

sents the couples that concurrently belong to the ref-

erence alignment and to the output of the alignment

method. The precision is the ratio of the number of

Ontology Alignment Evaluation Initiative - OAEI-2007

Campaign, http://oaei.ontologymatching.org/2007/

pertinent found couples, i.e., ”N

correct

”, by the number

of total couples, i.e., ”N

f ound

”. Thus, it represents the

part of the true correspondences between those found.

The precision function is deﬁned as:

precision =

| N

correct

| N

f ound

. (4)

The recall is the ratio of found pertinent couples,

”N

correct

”, by the total number of pertinent couples,

”N

expected

”. It speciﬁes, the part of the true found cor-

respondences. The recall function is deﬁned as:

recall =

| N

correct

| N

expected

. (5)

Precision and recall metrics are used to perform

the evaluation of OACAS method.

3.2 Experimentation and Results

The main objective of the experimentations with the

OACAS method is to ﬁnd the best combination of lin-

guistic measures. In the experimental study, various

measures have been used. The goal is to experiment

different measures in order to ﬁnd the more appropri-

ate measure associated to the node descriptors. In or-

der to achieve the objective, 27 arrangements of tests

have been experimented. Each test uses a particular

combination of similarity measures to compute lin-

guistic similarities between the descriptors of entities

to be aligned. During the process of the carried out

tests, different weights were assigned to the descrip-

tors (names, comments and labels). The nodes to be

aligned can have different descriptors. Depending on

the descriptors of the nodes, different weights are at-

tributed. In the case where the nodes are described by

three descriptors, the weights are 0.8, 0.1 and 0.1 as-

sociated respectively to the names, comments and la-

bels. Whereas the nodes contain only names and com-

ments descriptors, the weights are respectively 0.85

and 0.15. The weights 0.85 and 0.15 are assigned to

the names and labels where those the entities are de-

scribed by them. The experimental results obtained

are developed in the next subsection.

The combination using three different linguis-

tic similarities (LEVENSHTEIN, Q-GRAM and JARO-

WINKLER) is the best one. In fact, the LEVENSHTEIN

measure is more appropriate for computing linguistic

similarity between the names of entities to be aligned.

Whereas, the Q-GRAM measure is more indicated

to compute linguistic similarity between comments

of ontological entities. JARO-WINKLER measure is

more appropriated for computing linguistic similarity

between the labels of entities to be aligned. Indeed,

names and labels of ontological entities are short

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

236

strings. For this type of strings, LEVENSHTEIN and

JARO-WINKLER measures are more adapted to com-

pute the linguistic similarity. Comments are strings

composed with many words. For this type strings, the

Q-GRAM measure gives the best linguistic similarity

values. Columns 2-3 of Table 1 of the Appendix show

precision and recall values for the best combination.

The next subsection will compare OACAS vs other

related methods.

3.3 Comparative Study

In order to evaluate the results obtained by OACAS,

Tables 1 and 2 of the Appendix show the obtained pre-

cision and recall values. They also represent values

respectively obtained by FALCON-AO, ONTODNA

and RIMOM methods. With respect to Tables 1 and

2, the OACAS method produces better results than

the three others methods, in particular in the 30x fam-

ily tests. Theses tests represent real ontologies. The

OACAS method gives worse results in the 26x family

tests. In this family of tests, ontological entities do not

have properties (names and comments). Interestingly

enough, these ontological components are the main

factors in computing the alignment score using the

OACAS method. The experimental results also high-

light that the performances of the OACAS method are

highly linked to the different characteristics of onto-

logical components (names, comments and labels).

In the case where the descriptors of the entities are

dropped, the quality (precision and recall values) of

the alignment is degraded. For example, in the tests

257 and 260 the ontological entities are not described

neither by names nor by comments. In addition, rela-

tions and properties are absent. Whenever the name

and comments descriptors are not present, the values

of linguistic similarity measures are lowered and con-

sequently the value of the composed linguistic simi-

larity will follow the same tendency. Moreover, when

the considered ontologies to be aligned do not contain

relations nor properties, the values of neighborhood

similarities decrease. The local (linguistic similarity

composition and neighborhood similarity) similarity

computation in the OACAS method reduces the ag-

gregation similarity value. For this reason, precision

and recall values are degraded.

4 CONCLUSIONS

In this paper, we introduced a new alignment method

of OWL-DL ontologies. The new proposed method

OACAS, allows to exploit at most the informative

present within in an ontology described in OWL-DL.

The process of alignement in the OACAS method,

contains two phases: a local phase and a phase of

aggregation. The local phase allows to calculate the

linguistic similarity consisted as well as the neighbor-

hood similarity. This two similarities are combined

during the second phase to determine the aggregation

similarity. The results obtained by the method OA-

CAS are very encouraging, compared with the results

obtained by the methods FALCON-AO, ONTODNA

and RiMOM. The method OACAS shows more suc-

cessful results compared to the other methods in par-

ticular on the real ontologies (the family of the tests

30x). In order to improve the OACAS method, some

improvements can be brought. Indeed, the method

OACAS have to deal with ontologies of huge sizes.

The integration of the API WORDNET (Miller, 1995)

is necessary, to improve the values of the measures of

the linguistic similarity.

REFERENCES

Euzenat, J. (2001). Towards a principled approach to se-

mantic interoperability. In Proceedings of the IJ-

CAI Workshop on Ontology and Information Sharing,

pages 19–25, Seattle, US.

Euzenat, J., Mochol, M., Shvaiko, P., Stuckenschmidt, H.,

Svab, O., Svatek, V., van Hage, W. R., and Yatskevich,

M. (2006). Results of the ontology alignment evalua-

tion initiative 2006. In Proceedings of the First ESWC

2006 international workshop on ontology matching.

Euzenat, J. and Shvaiko, P. (2007). Ontology Matching.

Springer-Verlag, Heidelberg (DE).

Gruber, T. (1993). A translation approach to portable ontol-

ogy speciﬁcations. Knowledge Acquisition, 5(2):199–

220.

Hu, W., Zhao, Y., Li, D., Cheng, G., Wu, H., and Qu, Y.

(2007). Falcon-AO: results for OAEI 2007. In Pro-

ceedings of the 2

International Workshop on On-

tology Matching (OM-2007), pages 170–178, Busan,

Korea.

Kiu, C. and Lee, C. (2007). OntoDNA: ontology align-

ment results for OAEI 2007. In Proceedings of the 2

International Workshop on Ontology Matching (OM-

2007), pages 227–235, Busan, Korea.

Li, Y., Zhong, Q., Li, J., and Tang, J. (2007). Result of on-

tology alignment with RiMOM at OAEI’07. In Pro-

ceedings of the 2

International Workshop on On-

tology Matching (OM-2007), pages 196–205, Busan,

Korea.

Miller, G. A. (1995). WORDNET: a Lexical Database for

English. Communications of the ACM, 38(11):39–41.

Ukkonen, E. (1992). Approximate string-matching with q-

grams and maximal matches. Theoretical Computer

Science, 92(1):191–211.

OACAS - Ontologies Alignment using Composition and Aggregation of Similarities

237

Valtchev, P. (1999). Construction automatique de tax-

onomies pour l’aide la repr

esentation de connais-

sance par objets. Th

ese de doctorat, Universit

e de

Grenoble 1, France.

APPENDIX

Table 1: Precision and recall values for OACAS, FALCON-

AO, ONTODNA and RIMOM methods (Part 1).

Valtchev, P. (1999). Construction automatique de tax-

onomies pour l’aide la repr

esentation de connais-

sance par objets. Th

ese de doctorat, Universit

e de

Grenoble 1, France.

APPENDIX

Table 1: Precision and recall values for OACAS, FALCON-

AO, ONTODNA and RIMOM methods (Part 1).

OACAS ONTODNA

Tests Pre. Rec. Pre. Rec.

101 1.00 1.00 0.94 1.00

103 1.00 1.00 0.94 1.00

104 1.00 1.00 0.94 1.00

201 0.66 0.15 0.11 0.01

202 0.78 0.15 0.11 0.11

203 1.00 1.00 0.94 1.00

204 1.00 0.95 0.93 0.84

205 0.91 0.40 0.57 0.12

206 0.00 0.00 0.69 0.23

207 0.00 0.00 0.69 0.23

208 1.00 0.95 0.93 0.84

209 0.00 0.00 0.57 0.12

210 0.00 0.00 0.69 0.23

221 1.00 1.00 0.93 0.76

222 1.00 1.00 0.94 1.00

223 1.00 1.00 0.94 1.00

224 1.00 1.00 0.94 1.00

225 1.00 1.00 0.94 1.00

228 1.00 1.00 0.53 0.27

230 1.00 1.00 0.91 1.00

231 1.00 1.00 0.94 1.00

232 1.00 1.00 0.93 0.76

233 1.00 1.00 0.53 0.27

236 1.00 1.00 0.53 0.27

237 1.00 1.00 0.94 1.00

238 1.00 1.00 0.94 1.00

239 0.91 1.00 0.50 0.31

240 1.00 1.00 0.50 0.27

241 1.00 1.00 0.53 0.27

246 1.00 1.00 0.50 0.31

247 1.00 1.00 0.50 0.27

248 1.00 0.10 0.11 0.01

249 1.00 0.10 0.11 0.01

250 0.00 0.00 0.00 0.00

251 0.00 0.00 0.11 0.01

252 1.00 0.16 0.11 0.01

253 1.00 0.08 0.11 0.01

254 0.00 0.00 0.00 0.00

257 0.00 0.00 0.00 0.00

258 1.00 0.08 0.11 0.01

259 1.00 0.08 0.11 0.01

260 0.00 0.00 0.00 0.00

261 0.00 0.00 0.00 0.00

262 0.00 0.00 0.00 0.00

265 0.00 0.00 0.00 0.00

266 0.00 0.00 0.00 0.00

301 0.95 0.83 0.88 0.69

302 0.96 0.88 0.90 0.40

303 0.96 0.85 0.90 0.78

304 0.96 0.95 0.92 0.88

Table 2: Precision and recall values for OACAS, FALCON-

AO, ONTODNA and RIMOM methods (Part 2).

Table 2: Precision and recall values for OACAS, FALCON-

AO, ONTODNA and RIMOM methods (Part 2).

FALCON-AO RIMOM

Tests Pre. Rec. Pre. Rec.

101 1.00 1.00 1.00 1.00

103 1.00 1.00 1.00 1.00

104 1.00 1.00 1.00 1.00

201 1.00 0.95 1.00 1.00

202 0.87 0.87 1.00 0.80

203 1.00 1.00 1.00 0.88

204 0.98 0.98 1.00 1.00

205 1.00 0.98 1.00 0.99

206 1.00 0.93 1.00 0.99

207 0.98 0.91 1.00 0.99

208 1.00 1.00 0.98 0.86

209 0.79 0.78 1.00 0.84

210 0.81 0.80 0.99 0.85

221 1.00 1.00 1.00 1.00

222 1.00 1.00 1.00 1.00

223 1.00 1.00 1.00 1.00

224 1.00 0.99 1.00 0.99

225 1.00 1.00 1.00 1.00

228 1.00 1.00 1.00 1.00

230 0.94 1.00 0.94 1.00

231 1.00 1.00 1.00 1.00

232 1.00 0.99 1.00 0.99

233 1.00 1.00 1.00 1.00

236 1.00 1.00 1.00 1.00

237 1.00 0.99 1.00 0.99

238 1.00 0.99 1.00 0.99

239 1.00 1.00 1.00 1.00

240 1.00 1.00 1.00 1.00

241 1.00 1.00 1.00 1.00

246 1.00 1.00 1.00 1.00

247 1.00 1.00 1.00 1.00

248 0.85 0.84 0.99 0.78

249 0.87 0.87 1.00 0.79

250 1.00 0.27 1.00 0.55

251 0.56 0.56 0.76 0.58

252 0.71 0.71 0.85 0.70

253 0.85 0.84 0.99 0.77

254 1.00 0.27 1.00 0.27

257 1.00 0.27 1.00 0.55

258 0.54 0.54 0.76 0.57

259 0.70 0.70 0.85 0.69

260 1.00 0.31 0.93 0.45

261 0.89 0.24 1.00 0.27

262 1.00 0.27 1.00 0.27

265 1.00 0.31 0.93 0.45

266 0.89 0.24 1.00 0.27

301 0.91 0.82 0.75 0.67

302 0.90 0.58 0.72 0.65

303 0.77 0.76 0.45 0.86

304 0.96 0.93 0.90 0.97

KEOD 2009 - International Conference on Knowledge Engineering and Ontology Development

238