Ontology method construction for intelligent decision support systems Olga Burceva, Ludmila Aleksejeva Department of Modelling and Simulation, Riga Technical University, 1 Kalku Street, Riga, LV-1658, Latvia olgaa.burceva@gmail.com, l.aleksejeva@rtu.lv Abstract: The lessons learned from experiments with early rule-based systems were not only that domain knowledge should be represented in an explicit way (such as the one supported by an ontology), but also that the problem-solving behaviour of a system should be carved out in a separate component of the system. The difficulty of finding the best alternative is often complex to resolve, especially it increases when it is necessary to consider alternative by several qualitatively different criteria. Available at the enterprises many information systems and decision support systems (DSS) are not able to fully meet the needs of managers because usually they represent a set of disparate databases, and the information in such systems is represented by text, or in the form of various directories and is characterized by a set of independent, uncoordinated and an implicitly expressed conceptual description of the system. Due to this it is necessary to have DSS that would provide decision makers information about the functions and mutual relations of the structural components of the company for their effective management. There are different approaches, models, and knowledge definition language for this but more and more popular recently become ontology engineering. The main idea of our approach is to separate the decision making method (method ontology) from the database (domain ontology) in the construction of ontology-based decision support system. Here, the problem of constructing domain ontology and method ontology appears. This paper focuses on constructing only method ontology. Keywords: method ontology, ontology construction, multicriteria decision support, AHP Introduction The difficulty of finding the best alternative is often complex to resolve; especially it increases when it is necessary to consider alternatives by several qualitatively different criteria. Many information systems and decision support systems (DSS) available at the enterprises are not able to fully meet the needs of managers because they usually represent a set of disparate databases, and the information in such systems is stored as text or in the form of various directories and is characterized by a set of independent, uncoordinated and implicitly expressed conceptual descriptions of the system (Bolotova, 2012). Due to this, it is necessary to have DSS that would provide decision makers information about the functions and mutual relations of the structural components of the company for their effective management. There are different approaches, models, and knowledge definition languages for this but ontology engineering is recently becoming more and more popular. At the formal level ontology is a system containing certain concepts, properties of concepts, relationships between concepts, and additional limitations as are determined by axioms (Khakhalin, 2005). Ontological engineering includes: definition of the concepts in the ontology, guidance taxonomy, development of concepts and situations structures, determination of properties and values of these properties, procedures for output and transformations of situations. Ontologies provide a number of useful features for intelligent systems, as well as for knowledge representation in general for the knowledge engineering process. Ontologies perform an integrating function, providing a common conceptual basis in the decision-making processes and a common platform to bring together a variety of information systems. Decision support systems The main problem of the decision-making theory is the selection of one or more best objects (options, alternatives, etc.), ordering or ranking objects based on their properties, classification or sorting objects by the specified categories. The properties of these objects are characterized by many attributes or estimates on many criteria, the available quantitative and / or qualitative scale (Aleksejeva, 1998). Preferences of the decision-maker (DM) are a key factor of rational choice. DM formalizes preferences, setting characteristics of the researched problems and properties of the objects, comparing solutions, evaluating the quality of the selection. Preferences can be defined by binary relations, functions, decision rules that have logical, mathematical and verbal form. At the same time, solving the problem, a person can express his preferences consistently. There are a lot of methods for selecting the objects described by many quantitative and/or qualitative attributes (Aleksejeva, 1998, Saaty, 2011). The best options selection is carried out using a variety of methods of optimal choice based on the search of one extreme or the many features that characterize the performance or quality of the solution (Petrovsky, 2009 a ). In the methods of multi-criteria optimization, generalized criterion is usually given by the convolution of many private numeric criteria in the form of a "weighted sum". However, determination of baseline weight is a serious problem. http://aict.itf.llu.lv 48
To order objects in general or by multiple criteria, there are commonly used methods based on pairwise comparisons of objects. If there are many criteria and/or multiple decision makers, the resulting ordering of objects is constructed based on the pairwise comparisons of estimates of vectors representing the objects. In the methods of analytical hierarchy (Saaty, 2011; Petrovsky, 2009 b ) the options are ranked according to their priority, which is consistently evaluated by pairwise comparison of options, assessment criteria and participants to the global goal of the problem being solved. This method will be used in this work to show, how the method ontology can be constructed. The Analytic Hierarchy Process (AHP) The Analytic Hierarchy Process was devised by Thomas L. Saaty (1977) in the early seventies. It is a powerful and flexible tool for decision-making in complex multi-criteria problems. This method allows one to gather knowledge about a particular problem, to quantify subjective opinions and to force the comparison of alternatives in relation to established criteria (Saaty, 2011; Loranzo-Tello et al, 2008). The AHP method includes the following steps: Step 1: making the hierarchy. The problem should be defined in a hierarchical structure. The hierarchical structure is like an inverted tree. At the top should be the goal to be achieved, or the problem to be solved. The following are the parameters which affect the value of the final decision criteria. It should be noted that the criteria can be split into subcriteria. The next are alternatives to achieve the goal. For each of these alternatives it must be possible to determine the absolute and relative importance of each criterion. Thus, the hierarchy allows us to decompose a complex problem into parts, which allows us to understand the complexity and diversity of the upcoming elections (Saaty, 2011). Step 2: setting priorities. All the criteria by which we are going to compare the alternatives must be mutually compared. Step 3: comparison of the alternatives. With knowledge of the relative importance of each criterion, we can go to the comparison of alternatives for each criterion. Step 4: check for consistency. If the procedures described above are performed by a group of DM, it is important to use the average of the personal ratings. Step 5: making the final decision. With the results for the pairwise comparison of alternatives and the relative importance of the criteria, we can calculate the evaluation of each of the alternatives, which will give us a basis for making the final decision. Ontology engineering Ontology defines the common words and concepts used to describe and represent an area of knowledge, and so standardize the meaning. Ontologies are used by people, databases, applications that need to share domain information. Ontologies include computer usable definitions of basic concepts in the domain and the relationships among them. They encode knowledge in a domain and also knowledge that spans domains. So, they make that knowledge reusable (Rothenfluh et al, 1996). At the formal level ontology is a system containing certain concepts, properties of concepts, relationships between concepts, and additional limitations as are determined by axioms (Khakhalin, 2005). Ontological engineering includes: definition of the concepts in the ontology, guidance taxonomy, development of concepts and situations structures, determination of properties and values of these properties, procedures for output and transformations of situations. There are many reasons why the need of ontology development appears (Grechko, 2005): for knowledge sharing among people or software agents total understanding the structure of the data; for re-use of knowledge of the subject area; to turn assumptions in explicit connection or dependence; to separate domain knowledge from the operational knowledge; to analyze the domain knowledge. Ontological engineering denotes a set of design principles, development process and activities, supporting technologies, and systematic methodologies that facilitate ontology development and use its life cycle-design, implementation, evaluation, validation, maintenance, deployment, mapping, integrations, sharing, and reuse. Related works There are some definitions of methodology for building ontologies, again assuming manual approach. For instance, the methodology proposed in (Uschold et al, 1995) involves the following stages: identifying the purpose of the ontology (why to build it, how will it be used, range of the users), building the ontology, evaluation and documentation. The building of the ontology is further divided into three steps. The first is ontology capture, where key concepts and relationships are identified, a precise textual definition of them is written, terms to be used to refer to the concepts and relations are identified, the involved actors agree on the definitions and terms. The second step involves coding of the ontology to represent the defined conceptualization http://aict.itf.llu.lv 49
in some formal languages (committing to some meta-ontology, choosing a representation language and coding). The third step involves possible integration with existing ontologies. Traditionally, ontologies for a given domain are constructed manually using some sort of languages or representation and rely on manual extraction of common sense knowledge from various sources. Recently, several programs that support manual ontology construction have been developed, for example, METHONTOLOGY (Ferndndez et al, 1997) or Protégé (Youn et al, 2006). Protégé 2000 assumes that knowledge-based systems are usually very expensive to build and maintain because knowledge-based system development is done by a team including both developers and domain experts who may be less familiar with computer software. Protégé 2000 guides developers and domain experts through the process of system development. Developers can reuse domain ontologies and problem-solving methods with Protégé 2000, shortening the time for development and program maintenance. One domain ontology that solves different problems can be used in several applications, and different ontologies can use the same problem-solving methods (Youn et al, 2006). The PROTÉGÉ-II is used in (Rothenfluh et al, 1996). There is shown also how reusable domain and method ontologies are combined into the task-dependent application ontology. The construction of a knowledge-based system starts from a declarative description of the domain and of the problem-solving method: the domain and method ontologies. The developer merges these ontologies to produce an application ontology that is both domain- and method-specific. To generate a run-time system, PROTÉGÉ-II interprets the knowledge base created by the expert as input to the problem-solving method. According to (Liu et al, 2005), decision ontology can be designed and used to conceptualize the knowledge for decision making process. Ontology can be used for the model base design and model management. Ontology is divided into two parts, i.e. domain ontology and modelling ontology, where domain ontology shows the terminology for decision making, concepts and terms. Decision making ontology which has been built in (Kornyshova et al, 2005) also is representing decision making knowledge; it includes concepts, their properties and relationships. Therefore, using the decision ontology the domain knowledge and decision models become more sharable and reusable to users from different background and interoperable to other software agents (Liu et al, 2005). Suggested approach The main idea of our approach is to separate the decision making method (method ontology) from the database (domain ontology) in the construction of ontology-based decision support system. The method ontology describes domain-independent method concepts, in contrast to the domain ontology, which describes methodindependent domain concepts. Method ontologies are abstract descriptions of the inputs and outputs of the problem-solving method. The method ontology describes the knowledge requirements and the knowledge roles of a given problem-solving method (Rothenfluh et al, 1996). Knowledge about the given domain may lead the developer to make changes in a generic method through method configuration. A generic problem-solving method should be decomposable: it should be divisible into some sequence of subtasks, which in turn are solved by other methods (Rothenfluh et al, 1996). Tasks to be solved In this work we are building the ontology method for decision support algorithm AHP. Fig. 1 shows a flowchart of the decision making process construction as logical basis for ontology construction. The flow-chart represents all general steps of AHP method. Using the given flowchart with it steps, the method ontology will be constructed in the next chapter. http://aict.itf.llu.lv 50
The problem is defined no Alternatives and criteria are defined Are all decisions compatible? yes Hierarchical structure is built Pairwise comparison matrix is formed no no Are all the hierarchical levels compared? yes The common priorities are defined Calculation of the priorities for each criterion Does the result meet? yes Obtained a better alternative / structured set of alternatives Implementation of the suggested idea Fig. 1. Flow-chart of the decision making process in AHP. In this chapter a case study of the AHP method ontology construction is described. The first step of the AHP method is to define the problem and the main objective is to make the decision. A first concept of ontology hierarchy can be a problem class. The next step of the method is a hierarchy tree building where the root node is a problem, the intermediate levels are the criteria, and the lowest level contains the alternatives. In the hierarchy there are also clusters. Cluster is a group of nodes at the same levels which are subordinated to some other level the top of the cluster. The clusters are formed by placement of links between nodes. So, the hierarchy has clusters and levels, in which one there are nodes of hierarchy goal, criterion and alternative (Fig. 2). Problem Hierarchy Level of hierarchy Cluster Node of hierarhy Goal Criteria Alternative Fig. 2. Fragment of the method ontology. There can be many final results of the problem a set of alternatives, arranged alternatives, best alternative. Those all will be defined as the next nodes of the goal. To construct our method ontology, we will further take a http://aict.itf.llu.lv 51
simple task which we will attempt to resolve using AHP method and at the same time constructing the next concepts of ontology. We will take decision matrix with four alternatives and six attributes with numerical values (Table 1). The next step of the AHP method is evaluating the hierarchy. Once the hierarchy has been constructed, the building of pairwise comparison matrix for each level must be done. It is necessary pairwise compare all the criteria by which we are going to compare the alternatives. Table 1 Decision matrix X1 X2 X3 X4 X5 X6 A1 2.0 1.5 2.0 5.5 5 9 A2 2.5 2.7 1.8 6.5 3 5 A3 1.8 2.0 2.1 4.5 7 7 A4 2.2 1.8 2.0 5.0 5 5 In the beginning, it is necessary to get the criteria evaluation. We evaluate the criterion mutual influence, using the ratio of the relative importance of the nine-point scale where 1 equal importance, 2 very slight superiority, lightweight superiority. Suppose that X1 criterion is more important than the X2 criterion with very slight superiority. As the matrix is symmetric, then X2 is better than X2 in ½ times. In this way, all the criteria are compared with each other and the pairwise comparison matrix is built; it can be seen from Table 2. Taking into account the given step, the next concepts of our ontology in criteria branch can be pairwise comparison and then pairwise comparison has pairwise comparison matrix. Next the eigenvectors and vectors of the local priority should be calculated (Table 2). Also the index of agreement should be found. Table 2 Pairwise comparison matrix X1 X2 X3 X4 X5 X6 Eigenvectors Normalized vectors of the local priority X1 1 2 2 2 1 0.5 1.26 0.2 X2 0.5 1 1 1 0.5 0.333 0.66 0.1 X3 0.5 1 1 1 0.5 0.333 0.66 0.1 X4 0.5 1 1 1 0.5 0.333 0.66 0.1 X5 1 2 2 2 1 0.5 1.26 0.2 X6 2 3 3 3 2 1 2.18 0.3 Now it is necessary to obtain estimates of each alternative by each criterion. If you already have an objective assessment, then they are just issued, and normalized so that the amount is equal to 1. If the assessment is not objective, then, as the author (Saaty, 2011) writes, it is necessary to use pairwise comparison, similar to the previously discussed criteria. In this original task, an objective evaluation has already been given, which we also use. The following general priorities should be calculated. The result of this task is given in Table 3. The result of the task using AHP method X 1 X 2 X 3 X 4 X 5 X 6 Priority vector numerical value 0.2 0.1 0.1 0.1 0.2 0.3 Global priority A 1 0.235 0.188 0.253 0.256 0.250 0.346 0.271 A 2 0.294 0.338 0.228 0.302 0.150 0.192 0.233 A 3 0.212 0.250 0.266 0.209 0.350 0.269 0.266 Table 3 A 4 0.259 0.225 0.253 0.233 0.250 0.192 0.231 Considering this step we can define that alternative concept can have two branches: the first one is like criteria branch, the second one is selection of alternative, nodes evaluation and global priority. Using the previous discussed information, the ontology can be built as it is shown in Fig. 3. http://aict.itf.llu.lv 52
Set of alternatives Goal Arranged alternatives Best alternative Eigenvector Node of hierarhy Criteria Pairwise comparison Pairwise comparison Goal matrix Index of agreement Vector of the local priority Alternative Selection of an alternative Nodes evaluation Global priority Fig.3. AHP method ontology. Summary and future work In this paper we have reviewed ontology engineering and ontology method. Based on the background study the method ontology has been built for AHP algorithm for decision support. The developed ontology can be reviewed and completed with some other elements. Our future research will include validation of the developed ontology using some domain ontology. Also we will work in a way to bridge the gap between XML and ontology and how to get from ontology to a specific XML schema. Acknowledgements Thanks to Dr.habil.sc.comp., Professor Arkady Borisov from Riga Technical University for help and support. References Aleksejeva, L.Y., 1998. Principles of Design and Procedures of Functioning of Decision Support Systems in Organizational Systems. Summary of PhD Thesis, Riga, 24 p. Bolotova, L.S., 2012. Sistemy iskusstvennogo intellekta: Modeli i tehnologii, osnovannyye na znaniakh (Systems of Artificial Intelligence: Models and Technologies Based on Knowledge). Moscow, Finance and Statistics, 663 p. (in Russian) Ferndndez, M. etc., 1997. METHONTOLOGY: From Ontological Art Towards Ontological Engineering. AAAI Technical Report SS-97-06, pp. 33-40 Grechko, A.V., 2005. Ontologia metoda analiza iyerarhiy Saaty (Ontology of the Analytic Hierarchy Process of Saaty). Artificial intelligence 3, pp. 746 757. (in Russian) Khakhalin, G.K., 2005. Predmetnaja ontologia dlya ponimania tekstov geometricheskih zadach (The Domain Ontology under Geometrical Text Understanding). Moscow, Research Centre of Electronic Computers, 9. p. Available at: http://www.dialog-21.ru/digests/dialog2008/materials/html/khakhalin.htm, 24.01.2013. (in Russian) Kornyshova, Y., Deneckere, R., 2010. Decision-Making Ontology for Information Systems Engineering. ER'10 Proceedings of the 29th International Conference on Conceptual Modeling, pp. 104-107 Liu, O., Ma, J., 2005. A Web Services Approach to Model Management in DSS. PACIS 2005 Proceedings, 31 p. Loranzo-Tello, A., Gomez-Perez, A., 2004. Ontometric: A Method to Choose the Appropriate Ontology. Journal of Database Management (JDM) Volume 15, Issue 2., p.18 Petrovsky, A.B., 2009. Gruppovoye mnogokriterialnoye prinyatiye resheniy s nesovpodayushchimi predpochteniami (Group Multiple Criteria Decision Making with Distinct Preferences). Scientific statements Belgorod State University 12, pp. 152-160. (in Russian) Petrovskiy, A.B., 2009. Teorija prinyatiya resheniy (Decision Making Theory). Publishing Center Academy, 400 p. (in Russian) http://aict.itf.llu.lv 53
Rothenfluh, T.E. etc, 1996. Reusable Ontologies, Knowledge-Acquisition Tools, and Performance Systems: PROTÉGÉ-II Solutions to Sisyphus-2. International Journal of Human-Computer Studies, Volume 44, Issue 3-4, pp. 303-332 Saaty, T., 2011. Prinyatie resheniy pri zavisemostyakh i obratnikh svyazyakh: Analiticheskiye seti. (Decision- Making in Dependence and Feedback: The Analytic Network). LIBROCOM, 360 p. (in Russian) Uschold, M., King, M., 1995. Towards a Methodology for Building Ontologies. Workshop on Basic Ontological Issues in Knowledge Sharing, held in conjunction with IJCAI-95, 15 p. Youn, S., McLeod, D., 2006. Ontology development tools for ontology-based knowledge management. Nonpublished research report, 100 p. Available at: http://research.create.usc.edu/cgi/viewcontent.cgi?article=1104&context=nonpublished_reports, 24.01.2013 http://aict.itf.llu.lv 54