Ähnlichkeit im Kontext: Verbesserung semantischer Ähnlichkeitsranglisten durch Kontextinformationen

点击翻译按钮获取中文摘要

Categorisation is a fundamental step in knowledge representation because it structures modelled knowledge into coherent graded groups of mutually similar members and relates these categories within a hierarchical and meteorological structure. This is a crucial task for information extraction and also the basic step that enables reasoning within an ontology or knowledge base. Semantic similarity measurements are used to calculate semantic distances between category members, but also across categories. Similarity measurement plays an important role in discovery and information extraction in heterogeneous sources where pure subsumption reasoning is too rigid and potential matches are not visible for the user. Current mathematical approaches to calculate semantic similarity between geospatial entity classes are only able to catch the similarity between feature-based categories, but cannot be used for role-governed and ad hoc categories. These types of categories group members that satisfy a certain relationship (e.g. transport(X,Y)) or play a spatiotemporal role in a given situation without having features in common which make them similar. For example, Human and Water are members of a flooding category because of their spatiotemporal role within an emergency scenario. Contrary to feature-based categories the relevant aspect for membership is not intrinsic but extrinsic. It is the named edge (transport) between nodes (X,Y). The goal of this project is to develop a formal framework for representing role-governed categories in onthlogies and to propose a mathematical semantic similarity measure (called role-based similarity measurement] to calculate distances between and within them. Its integration in geographic information services will lead to more efficient system support for people in decision situations. Information discovery fundamentally relies on similarity because it is often not precisely known what one is looking for. In contrast to subsumption reasoning based discovery it offers a ranking of the results. Similarity also supports identity assumptions such as needed in the area of cultural heritage. Two place names represent the same real world place if both share common events that are related (via our knowledge from historic documents) to them. Similarity measurements are further used in location based decision support services and allow for the integration of individual user profiles.