Millions of hours of audiovisual content are held by collections of dedicated broadcast, lm and sound archives, institutional or corporate archives, libraries and museums. There is a large heterogeneity between the di erent audiovisual archives resulting from their history and tradition but also from cultural differences of the countries where those archives reside. Consequently metadata models covering the work ows and necessities in the archives di er as well. This fact and the need for metadata for various di erent use cases in the archives lead to a number of metadata models and standards. Thus mapping between di erent metadata models is inevitable in practical applications. We are currently developing a system for automating metadata mapping by formalizing semantics of properties in the di erent formats and their relations [4], based on an intermediate ontology, namely the meon ontology [3]. In order to enable users to validate the automatically determined mappings visualization functionalities are required in the system. This paper describes the integration of the ontology visualization developed in [6] into our mapping system prototype. Creating comprehensive, clear and intuitive visualizations of ontologies and RDF graphs is an ongoing challenge. Di erent approaches can be found in applications for Semantic Web engineers. An example is Protege1, which is an open, platform independent environment for creating and editing ontologies and knowledge bases. The application is extensible by its plug-in architecture and thus provides several visualizations. IsaViz2 is a visual tool for browsing and authoring of RDF models. Resource nodes are represented by ellipses, literals as rectangles and properties are displayed as lines with arrows. OntoSphere 3D [1] uses a collection of three-dimensional visualization techniques displaying ontologies. gFacet [2] combines the graph visualization with facet search in the graph. These applications use di erent kinds of visualization techniques to present the user a possibly easy to understand and complete overview of the whole RDF graph. Using graph visualizations of RDF data especially for end users has a number of drawbacks. For example, these visualizations are at and every node is treated as a primary node. Also, displaying a graph with hundreds of nodes and edges results in a cluttered visualization (cf. [5]). Nonetheless graph visualizations have their place, especially for Semantic Web engineers [6].
The prototype3 helps users nding, validating, and understanding metadata mappings by automatic metadata matching and appropriately visualizing mapping relations. Figure 1 shows the textual part of the user interface where the user creates a query. The rst step of the user is to select an input- and output metadata format. Via the Load button, the application lists all available concepts from the selected formats. The next step is to select one or more concepts for which the mapping relations should be found. After con rming the selected concepts by clicking the Ask button, dependencies between the input and output concepts according to the de ned rules are calculated and displayed. For each selected output format the information whether a mapping is feasible or not is displayed: True, if the output concept can be mapped from one or more of the selected input concepts, False otherwise. In case that there are output concepts without corresponding selected input concepts, the Find requirements option can be used. After selecting this option additional necessary input concepts are computed in order to establish mapping relations to the selected output concepts.
In addition to the boolean information about the feasibility of the mapping explained above, possible mapping relations are visualized in a graph visualization. The RDF-like graphs include the selected input concepts as yellow nodes, the selected output concepts as red nodes, the related meon concepts as green nodes and potentially missing input concepts as white nodes. It focuses on the current task of the user by displaying only necessary nodes and edges. This kind of visualization supports the user in understanding and validating the found mapping relations between input and output concepts of the metadata formats. An example of the graph visualization of the selected concepts shown in Figure 1 is depicted in Figure 2. The graph representation reveals that the input concept mpeg7:Height together with mpeg7:Width can be mapped to the selected output concept ma:FrameSize via meon:Resolution, which is part of the intermediate ontology (meon ontology). Beside this positive mapping relation, no appropriate mapping relation can be established to the remaining output concept ma:Creator from any of the selected input concepts. However, mpeg7:UnqualifiedCreator is a possible input concept to map to ma:Creator.
The mapping prototype is a Web application using standard Web technologies such as HTML and JavaScript for the user interface as well as Scalable Vector Graphics (SVG) for the graph visualization. To generate the graph we use the Java Universal Network / Graph Framework (JUNG)4, which provides a number of layout algorithms and mechanisms to manipulate graphs. An internal evaluation has shown that the \self-organizing map layout for graphs" produces the best results for our requirements. However, this layout algorithm generates a di erent layout at every single run. Therefore, it is necessary to animate the graph visualization for the user. The animation helps the user to follow how
the layout changed since the last run. For processing the RDF data the Jena Semantic Web framework5 has been used. 3
In this paper we have presented the visualization functionality of our metadata mapping prototype6 which helps users nding, understanding and validating metadata mappings by automatic metadata matching and appropriately visualizing mapping relations. The visualization shows mapping relations between input and output metadata formats which are determined by the system via an intermediate ontology. It uses coloured nodes and focuses on the current user task by displaying only nodes which are necessary for the current task. The system is able to nd direct metadata mappings as well as to suggest further input concepts to satisfy the desired metadata mappings. In the future the system shall support the de nition of mapping rules by the user in order to improve the results in cases where incomplete or ambiguous mappings between pairs of formats exist.
The research leading to these results has received funding from the European Union's Seventh Framework Programme under grant agreement nr. FP7 231161, \PrestoPRIME" (http://www.prestoprime.eu).