Multi-context based browsing in Heterogeneous Semantic Spaces: The Ontology Switching Approach

Homogenization has been the most widely applied approach to create common semantic spaces when it is necessary to overcome the limitations of sharing knowledge posed by semantically heterogeneous concept structures. Ontology switching manages diverse ontologies by integrating them and enables the user to access them in parallel. Most important, the user can change his point of view by switching to another concept structure at any point and explore different viewpoints according to his current context. An ontology switching system dynamically reorganizes the concept structure and populates it with knowledge objects. The paper describes the design options for ontology switching systems, presents prototypes, evaluation results and points to future developments. This work integrates research on the semantic web and on humancomputer interaction. 1 From Semantic Heterogeneity to Ontology Switching Ontologies are structured collections of concepts which describe the world. Ontologies define a perspective on the domain under consideration and have been an important tool for knowledge management. The disparity of ontologies has often been discussed as a major problem for mutual understanding and the sharing of knowledge. Most knowledge domains have been organized into ontologies several times form different perspectives. Each perspective is justified within its own context. Ontologies also differ over time. An example is the international patent classification system (IPC) which needs to be adapted to new scientific and technological developments in order to integrate new innovations. Consequently, each version of the IPC represents a description of the technology at one point in time and none of the older version is wrong. Furthermore, all versions are valuable for the user of patents because older assignments of patents to classes are not reviewed and updated after a change of the classification system. 1 http://www.wipo.int/classifications/ipc/en/index.html As a consequence, information system engineers are faced with a large variety of ontologies for each knowledge domain. Sharing knowledge across different perspectives represented by semantically heterogeneous ontologies remains tedious. Our ontology switching approach does not try to unify the semantic representations by fusing the heterogeneous ontologies. Much rather it integrates all relevant ontologies into one user interface and allows the user to chose his current perspective. That means, the user can change the viewpoint form which he explores the knowledge objects. The user selection can be changed at any point in the system. Switching causes a dynamic reorganization of the objects according to the selected perspective. The switching process invokes a process which display the new concepts, populates them with the knowledge objects and sets the current focus based on the position in the previous ontology. This can be done either based on a class or an object level. In the first case the user is presented with a view on the class from the selected ontology most similar to the one he had viewed in the previous ontology. In the second case, the same object is still shown to the user, however, it has been dynamically introduced into the selected ontology. The remainder of the paper is organized as follows. The second section discusses the heterogeneity of ontologies. Section three presents the ontology switching approach as a theoretical concept. Section four shows the implementations of the concept in information systems. The outlook points to areas of further development. 2 The Ontology Switching Approach The central idea of ontology switching is that the user should be enabled to change the ontology with which he wants to explore the knowledge objects in a unified collection. By allowing that, the user can select the appropriate perspective for his current context. Enabling ontology switching within each concept is motivated by principles from human-computer interaction. For the interaction, parameters once set by the user should be kept and the user should not be forced to enter them again. Within ontology switching, the parameter setting is done by selecting concepts thus expressing interest for them. Simply changing the underlying ordering structure by selecting another ontology should not invalidate the expressed interest and force the user to search for the concept of his interest in the newly selected ontology again. Ontology switching [Mandl & Womser-Hacker 2003] deals with heterogeneity in several levels. It integrates several overlapping collections indexed with different ontologies. Ontology switching provides a browsing system with a user interface containing all classifications. Between the ontology entries and the documents, relations need to be established where they cannot be found in the original system. For example, when a book is only available in one library, we need to index this book with terms from other library catalogues as well. This can be done manually, semiautomatic or fully automatic. Ontology switching can integrate the perspective of different departments within a company. An engineer, an auditor and a manager all take a different approach toward documents available to them. • Ontology switching results in the following value added services: • One browsing user interface serves for several ontologies • The reach of each ontology is increased • Ontology switching is made possible and the user can change his perspective • Ontology switching is based on the decision of the user and not on system limitations • Thematic selections remain effective during switching and are not lost Ontology switching is an application based on ontology learning. It exploits knowledge engineering and the results of the assignment of objects to ontology concepts. That way, ontology switching systems, create transparency in semantic heterogeneity. Ontology switching also supports ontology learning. Continuous use of Ontology switching systems allow the identification of highly used ontologies as well as fractions of ontologies which are accessed more than others. These popularity measures can be exploited in further ontology refinement. Fig. 1: MyShelf as an integrated information system Multiple assignments are also possible within ontology switching. Often, machine learning systems assign vague membership values to objects. These vague assignments can be integrated in several ways into ontology switching systems: • Complete ranked list within a concept • Object is only assigned to the concept where it receives its maximum membership value • Combinations between the two approaches above M yShelf Brow singInterface Information System