Ontology Mapping using Background Knowledge Zharko Aleksovski Philips Research Eindhoven Vrije Universiteit Amsterdam De Boelelaan 1081a Amsterdam, The Netherlands zharko@few.vu.nl Michel Klein Vrije Universiteit De Boelelaan 1081a Amsterdam, The Netherlands michel.klein@cs.vu.nl ABSTRACT In this paper, we report on a method for aligning two lists of terms using structure-rich ontologies as back- ground knowledge. The results of the method can be seen as suggested mapping candidates to users that perform an ontology alignment task. We applied the method on lists of medical terms, and we discuss the outcome. Categories and Subject Descriptors: J.3 [Life and Medical Sciences]: Medical information systems I.2.4 [Knowledge Representation Formalisms and Methods]: Representation languages E.1 [Data Structures]: Graphs and Networks General Terms: Algorithms, Theory. Keywords: ontology, semantic web, ontology align- ment. 1. INTRODUCTION There already exist several techniques for ontology align- ment [4, 1, 3, 2]. However, the existing techniques are most effective when the ontology contains structure. If the vocabularies have no structure, and only consist of lists, these approaches fall back on lexical techniques only. In this paper, we present a method for aligning two lists of concepts using other structure-rich ontolo- gies as a background knowledge. 2. MATCHING LISTS OF CONCEPTS US- ING BACKGROUND KNOWLEDGE We use two methods to match two lists of concepts. We assume the concepts to have labels that contain the meaning of the concepts in natural language. The first method we call lexical matching, the second semantic matching. In the second we use background knowledge. Copyright is held by the author/owner(s). K-CAP’05, October 2–5, 2005, Banff, Alberta, Canada. ACM 1-59593-163-5/05/0010 Figure 1: Comparing two labels: “Long brain tumor” and “Long tumor”. The first is more specific than the second because it consists of a superset of the words from the second label. 2.1 Lexical matching of concepts Lexical matching of concepts is a simple heuristic that only makes use of the labels. If one concept label is derived from another by adding extra words, then the first concept is considered to be a subclass of the second. An example is depicted in Figure 1. 2.2 Semantic matching on properties using back- ground knowledge In a first step, we establish relationships between the concepts in the two lists to be mapped and a struc- tured source of background knowledge. Through this relationship, the concepts in the lists acquire proper- ties with values taken from the background knowledge. From this background knowledge we can induce a map- ping between the concepts in the list by the heuristic that if two concepts in the lists have common proper- ties, and if they have related values for these properties, then these two concepts are related. In the medical do- main, as shown on Figure 2, a concept that describes a disease has a property “anatomical location”. If one disease has location “artery”, and another “aorta”, then these two diseases are related because according to the ontology “aorta” is kind of “artery”. 3. EXPERIMENTS In the experiments we used data from the medical do- main. We performed tests to match two lists of concepts that denote reasons for admission (RfA) in the intensive care unit - that is why a patient was brought in the IC. 3.1 Description of the test data The first list contains 1399 RfA. It was developed at the OLVG hospital in Amsterdam. The second list of