MappingAssistant: Interactive Conflict-Resolution for Data Integration

Enterprise applications often face the problem of integrating heterogeneous data due to the growing number of distributed systems. Leveraging semantic web technologies with ontologies as target schema for matching data is a successful approach for data integration. We developed a new interactive approach for identifying errors within alignments in the scope of the MappingAssistant project. It is based on a diagnostic method combined with human-understandable explanations on the instance level. This supports the users in finding erroneous rules or facts in a time-saving manner. Introduction. Alignments produced by automated ontology matching algorithms are error-prone. Consequently, their results need to be supervised by a human domain expert. As they are usually represented in ways only technical experts can deal with, the evaluation task is complicated and time-consuming. The MappingAssistant simplifies the alignment evaluation process by decreasing the amount of user interactions needed to correct the automatically generated mapping results. In particular, we developed a user-friendly conflict resolution method utilizing data reduction techniques for illustrating representative individuals and an algorithm for minimizing user interaction in the evaluation process of aligned data. User Interface. In data integration scenarios users are often faced with illlabeled concepts as for instance cryptographic database labels. As a result, the expert is used to investigate the data on the instance level. Existing applications like the AgreementMaker [2] illustrate alignments mostly on the schema level. Our framework also represents alignments on the concept level through a graph-based view and a table-based view (Figure 1 upper left and upper right). Additionally, we visualize instances (Figure 1 lower right) and natural language based presentations of proof-tree nodes (Figure 1 lower left). The user is able to specify the correctness of the matching either on the concept level, the instance level, or through the help of our diagnostic-based proof-tree dialog. The approach is completely implemented as an extension of the OntoStudio [1] environment.