Towards a tool for ontology engineering

TOWARDS A TOOL FOR ONTOLOGY ENGINEERING' J.A.ALONSO-JIM~NEZ, J.BORREGO-D~AZ, A.CWAVEZ-GONZALEZ Dept. of Computer Science and Artificial Intelligence-University of Sevilla {jalonso, jborrego, tchavez}Ous.es ABSTRACT A tool based on a spatial representation of provisional ontologies is designed. The tool allows the cleming of Knowledge Bases, aa well to induce new concepts in early steps of the building of an ontology. KEYWORDS: Qualitative-Reasoning, Knowledge-Based System, Ontology Engineering 1 INTRODUCTION Ontologies provide us with a common undemtanding in fields a Knowledge Management and +commerce [I]. FOF a satisfactory transition kom the actual WWW to the Semantic Web it will be need to deal with evolving ontologies, because of the key role they play in the reasoning services €or Knowledge Bases (KB) in the Semantic Web [Z] [l]. The aim of this paper is to show the foundational issues for ti tool to repair KB in provisional ontologies. It is based on a spatial representation of incomplete specifications of the concepts of the ontology using RCC [3], a sound theory for qualitative spatial reasoning. Then we use two types of actions, topological and reticular arrangements on the spatial representation, in order to repair some anomalim. Additional steps require the interaction with the belieh of the mer. This work is hmed on the analysis given in [4] on RCC knowledge databases. We aim to analyze a certain type of anomalies that arose from a eleaningeycle applied to KB's associated to complex ontologies [6]. The advantage of using this semantics lies in the fact that it is necessary to iniprove the current data cfeaning systems with a clear separation between the logical specification of data transformation and their physicd transformation, an explanation of the reasoning behind cleaninE results, and the possibility of interactive facilities to tune data cleaning programs. Description Logic (DL) is used to represent metadata. DL is a sound formatism to give a clear semantics to several tools for Knowledge Representation (KR) (8ee.e.g. [Z] Formally, Description Logics (http: //dl .kr. org) is R subset offirst urder logic. Thus it inherits aformalixed semantics, as opposite to some early formalisms for KR. DL deals with the representation of structured concepts, describing them with a language with specific features, as conjunction, quantifiers on attributes of concepts, etc. A KB, E, in DL is a pair (7, A), where A is a set of facts (the eztensional component) and T is a set of relations between concepts (the intensionel component). In fig. 1, a little KB on the family ontology is given, which will be our running example. 2 ANOMALIES IN PROVISIONAL ONTOLOGIES Knowledge Bases in DL may be affected by classical anomalies. We must bear in mind 'Work partially supported by the MCyT project TIC 2000-1368-CO3-0 and the project TIC-I37 of the Pl~n An&lw de hmtigacrdn. [51). 193 Authorized licensed use limited to: Universidad de Sevilla. Downloaded on January 20,2021 at 18:33:48 UTC from IEEE Xplore. Restrictions apply. Father(John) Man(John) Woman(Ann) 3harChild. Pew (Ann) Father& Man n jhasChild,pelMn A Fema'e(RuPau') Mam(RuPau') Woman E Person fl Female Man C Person n -Woman Fatha Parent Figure I: A pmvisional KE, C = (7, A), on the family ontology Nx, Y) ft -C(x, Y) P(x, Y) tf W(Z, x) 3 c(z, Y)1 PPb, Y) +) P(x, Y) /I +(Y, x) EQ{x, Y) ff P(x, Y) A P(Y, x) Ob, Y) gz[P(z, x) A P(z,Y)] (x overlaps y) OR(% Y) ft -O(x, Y) Po(x, Y) ft O(x, y) A +(x, Y) A +(Y, x) TPP(x, y) H PP(x, y) A 3z{EC(z, I) A EC(z, y)] NTpP(x, y) H PP(x, y) A -3z[EC(z,x) A EC(z, y)] (x is disconnect from y) (x is part of y) (x is proper part of y) (x is identical with y) (x is discrete from y) (x partially overlaps y) (x is externally connected to y) (x is a tangential prop. part of y) (x is a non-tang. prop. part of y) Y) tf Cb, Y) A -ob7 Y) Figure 2: Asionas of RCC the possible dynamic natiire of ontologies, and in the first phases of their building they must be considered as provisional Even if an ontology lacked of unacceptable anomalies, there would be several versions we must work on. Incompleteness of a KB must be understood in two ways: the logical incompleteness ( with respect to a kind of queries), ttnd due to the lack of concepts or roles (incompleteness with exprwsive nature). There are anomalies due to Iwk of an exact profile of several concepts. When it occurs, the user works on beliefs not even explicited in the KB. Such concepts will be called notions. The existmm of notions in an ontology implies that two concepts covered by the same notion cau not be distinguished, namely the ontology is mrse. On the other hand, neglected development of the ontology leads to a problem in the validation field, different from classical validation task in Knowledge Based Systems: the ontology dow not fit in with the user beliefs about his/her huework, or it is both hard to use and to be understood by the others. Messy ontologies definitively are a risk for the management of large KB in the Semantic Web. 3 SPATIAL REPRESENTATION OF ONTOLOGIES The Region Connection Calculus (RCC) [3] is a topological approach to qualitative spatial representation and reasoning on spatial entities, which are non-empty regular seta. The basic relation between them is the connection relation C(x, J), whieh is interpreted as "the closures of x and y tnnberseet". The axioms of RCC are two basic axioms on C, A, := Vr[C(x,x)] and A2 := Vx,y[C{x,y) -+ C(y,x)], plus several axioms/definitions on the main spatial relationships, we fig. 2. The eight jointly exhaustive and pairwise disjoint relations forming the relational calculus RCC-6, have been deeply studied by J. Renz and B. Nebel [8]. We will use later two kinds of motions on this relational calculus: reticular motions and topologird motions. They are cognitively adequate motions and have continuous nature. Reticular motions are refinements of relationships (downward motions in the 194 Authorized licensed use limited to: Universidad de Sevilla. Downloaded on January 20,2021 at 18:33:48 UTC from IEEE Xplore. Restrictions apply. Figure 3: The ontology transformation pmcess Figure 4 Initial gmph (left) and aolufton (right) lattice of the RCC relationships). Topological motions are motions of least topoIogicul distance, as the substitution of a relation by other one cognitively nmr. It will be used too the reticular projection on RCC-8, R I+ R := {R' E RCCS : R' L R}. The initial KB is a DL knowledge base. The cleaning process is shown in figure 3. The result of the process will be a new K3 consistent with the beliefs of the user. Indeed, the process must he a cycle, because it is possible that the ontology changes, new data have been induced, and they lead to a new revision. 4 FIRST STEP: GRAPHICAL INTERPRETATION Firstly, a constraint satisfaction problem (CSP) on the spatial relational calculus RCCS (or RCCS) is produced by a cognitively sound translation of the TBox to RCC formulas, obtaining a consistent scenario, which i represented in 2D. Facta of the Abox are added as points. Each elementary concept A of C is interpreted as a region in the phne, A' C R2 (we will write A' = A). In order to carry out this interpretation, a translation of C to RCCS is applied, translating the formulas of TBox to a set of RCC formulae as follows: (C C D)" = {F(C, D)}, (C D1 R D2)' = {P(G, DL). P(C, 02)}, (C C 01 U D2)* = {O(C, Q), O(C, Dz)}. (Notice that in this case, if C n DI = 0, the u9er will discard O(C, DI) Iater). Each fact of the Abox is trans1ated as A(a) ct a E A. For our example, the graph of constraints is given in fig. 4 (left), if we make reticular projection on RCCS. A solution is on the right in fig. 4. The consistent scenario 191 is spatialIy represented by a set of (not necemarily connected) regular regions (see flg 5 left). 195 Authorized licensed use limited to: Universidad de Sevilla. Downloaded on January 20,2021 at 18:33:48 UTC from IEEE Xplore. Restrictions apply. I I i U"&" I Figure 5: Spatial representation of the solution (left) and after the arrangements (right) 11 3haaChild.P ] Figure 6: Spatial relationships among the concepts 5 SECOND STEP: SPATIAL ARRANGEMENTS In this step the user is requested to make reticular and/or topological arrangements on the graphical representation. By introduction of new regions, new concepts might be introduced too. For our example, the new picture is on the right of fig. 5. 6 THIRD STEP: A NEW KB When the user believes that the current spatial Bcenaxio is a sound representation, we must translate it to a new KB. It is necessary to make some remarks on the spatial scenario. The spatial relationships may be inadequate with respect to the mental ontoiogy believed by the user. This anomaly is detected when the user rebuses the translated KB, producing a new graphical refinement. Actually, the above translation must be applied to a representation of the map by a graph that the user thinks as fair. Next we define the translation R E RCC ++ R' of each relation on RCC to a set of DL formulae by recursion in the order of the RCC axiomatization (fig. 2). The relation Q E A is translated to A(a). From now, "element" means "spatial interpretation of a constant symbol". One selected translation of rule are: PP(A, B)' = P(A, B)* if it exists a region D such that C(B, D) A -C(A,D). In other case, a new concept name is introduced NB/A, and PP(A, 5)' = P(A, D)* U {Nq, C 8, Ne\, E TA}. The names far the new concepts does not refer to any intended feature of the concept. For example, it is not initially true that N~\A E B \ A. For example, the table of relation8 is in fig. 6 and the KI3 obtained is shown in 7. 7 FOURTH STEP: EVALUATION BY THE USER One of the goals of this step is to give a name for new concepts. This implies that the u8er must make an effort for interpreting the resiilt. Morewer, the user must decide if 1 96 Authorized licensed use limited to: Universidad de Sevilla. Downloaded on January 20,2021 at 18:33:48 UTC from IEEE Xplore. Restrictions apply. Father E Parent r Man fl -Female Female g Pew Woman E Person Man 5 Person Partnt L Perron Man E -Woman 3harChild.Person Parent Crossdresser Female Crosdmser E -Woman Mother Parent n Woman A" = A' U { MOther(Ann) Crossdresser(Rupaul) Figtire 7: KE f" spatial repmsentation and KB after the lad step the elements belong to the new concepts, if they are topologically close. For example, the user must decide if Rtlplnul E NFemole\~oman. It is possible that a new concept had been discarded. It might occur if the graphic representation the user has done became inadequate for her/his beliefs In our case, the user gives a name to the expression NF~~~~~\w~~~ that is Crossdresser; the expression Npannt\fpther is named Mother, and it is identified with N3hasChild.perron\Father. The final KB is on figure 7. If the user had believed that DR(Femde, Man), the translation would produce the concept name AbFemale,Man and the fact AbFe,,l,,~,,(RuPaut) would be included into the Abox, but no relation between AbFemalt,Man and Man or Female is added. It is preferable to be still a notion. 8 CONCLUSIONS, RELATED AND FUTURE WORK We showed how to translate, with plausibility cognitive, theanalysis of KB to agraphical refinement with a sound qualitative spatial reasoning tool, RCC. The representational principle on we work is that. an acceptable t;mall set of concepts must have a clear spatial representation. In other case, this set should he messy. This hypothesis is mgud by the practice. The tool is not for reasoning service based on entdment. Nonmonotonic reasoning (as default) is uzed in several steps of the cleaning process. The tool is more related with the graphical representation of ontologies and the mereological analysis of concepts 197 Authorized licensed use limited to: Universidad de Sevilla. Downloaded on January 20,2021 at 18:33:48 UTC from IEEE Xplore. Restrictions apply. given in [lo]. In fact, our reprewntation satisfies, in a non temporal setting, the minimal requirements proposed in the cited paper, There are related works on reasoning about concepts such as Galois lattices. In [ll], information on visual tools to represent concepts lattices are given, but their aim is not specifically to transform the ontology because it is supported by real data. On the other hand, the method allows us to use logically consistent reasoning for repairing ontologies. There exist two research line we are currently studying. First, although in this paper we do not deal with a spatial representation of the roles of KB, this feature can be added to the tool. On the other hand, the tool may be considered aa a learning process (based on generalization and refinement of concepts). Thus it seems interesting to design metrics reflecting such a process by means of adapting techniques for clauaal learning [12]. References D. Fcnsel: Ontologies: Silver Bullet for Knowlcdge Management and Electronic Comnierce. Springer-Vcrlag (2000). [. Horrocks, D.L. McGuinnes, C. A. Welty. Digital Libraries and Web-Based Information Systcms. In F. Baader et aL (ed.) The Description Logic Handbook. pp. 436-459. Cambridge University Press (2003). A. G. Cohn, E. Bennett, J. M. Gooday and N. IM. Gotts. Flepresenting and Reasoning with Qualitative Spatial Relations about Regions. In 0. Stock (ed.) Temporal and Spatial Reasoning, pp 97-134. Kluwcr (1997). J.A. Alonsdimhez, J. BorrepDiaz, Antania M. Chbvcz-Gonzalea, M. A. GutiBrraNaranjo and Jorge D. Navarrc-Marin. Towards a Practical Argumentative Reasoning with Qualitative Spatial Databases. Dcvelopments in Applied Artificial Intelligence, (IEA/AIE 2003) (LNAI 2718) Springer-Verlag. pp.789-798 (2003). G. Antoniou and F. van Harmelen. Web Ontology Language: OWL. In S. Staab and R. Studer (eds.) Handbook on Ontologies in Information Systems, pp.67-92, Springer-Verlag (2003)~oc. of 27th Very Large Databases Conference VLDB. Morgan Kauffman (2001). J.A. AlonseJirnBnn, J. BorregsDiaz, Antoilia hl. ChLve&onz&lez, M. A. GutiBrrezNaranjo and Jorge D. Navarr+Marfn. A Methodology for the Computer-Aided Cleaning of Complex Knowledge Databases. hoc. of 29th Annual Conference of the TEEE Industrial Electronics Society, IECON 2002. IEEE. Press. pp. 1806-1811 (2002). 8. Galhardas, D. Florescu, D. Sasha, E. Simon, C.A. Saita. Declarative Data Cleaning: Language, Model and Atgorithms. Proc. of 27th Very Large Databases Conference VLDB. Morgan Kauffinan (2001). 3. Renz, B. Nebel. On the Complexity of Qualitative Spatial Reasoning: A Maximal Tkactable Fragment of the Region Connection Calculus. Artificial Intelligence lOS(1-2): 69-123 (1999). J. Renz. A Canonical Model of the Region Connection Calculus. Proc. of the 6th Int. Conf. on Principles of Knowledge Fiepresentation and Reasoning (KR'98) (1998). pr'. Guarino and C Wekty. Towards a Methodology for Ontology-Based Model Engineering. ECOOP-2000 Workshop on Model Enginwring (2000). B. Ganter, R. Wille. Formal Concept Analysis-Mathematical Foundations. Springer M.A. Gutihrrez-Naranjo, J. A. Alons+JimBnez and 3. Borrego-Diu. A Quasi-metric for Machine Learning. Advances in Artificial InteUigence-13ERAMIA 2002. Lecture Notes in Artificial Intelligence 2527, pp. 193-203 (2002). (1999). 198 Authorized licensed use limited to: Universidad de Sevilla. Downloaded on January 20,2021 at 18:33:48 UTC from IEEE Xplore. Restrictions apply.