A half-day workshop held at ICCBR 2010, Alessandria, Italy, July 19-22, 2010
Provenance has emerged as a burgeoning research area in scientific computing and the semantic Web. Case-based reasoning systems exploit the results of prior reasoning, by storing cases for reuse. However, comparatively little attention has been devoted to how CBR can exploit knowledge of the provenance of stored cases. Such knowledge may include information about external sources of cases and the context in which they were captured, derivation traces of cases and other meta-data. For example, case provenance information can be used to assess trust in sources and confidence in related cases, or for metalearning to determine the effectiveness of the various adaptation strategies which produced a system’s solutions, in order to improve future reasoning. This workshop investigates the interplay of case provenance with areas such as trust and reputation, reasoning and meta-reasoning, and explanation.
Many domains offer opportunities for capture of provenance and meta-data useful for CBR applications. For example, in e-science, interest in the collection of provenance information from the execution of scientific workflows provides a knowledge source for CBR applications within this domain. In e-commerce, information about communities of recommenders may provide useful information for assessing and applying their recommendations. In help-desk domains, information about the context in which faults occur (e.g., available technicians and client locations) may provide important clues for fault diagnosis and recovery.
Likewise, the CBR cycle itself provides numerous opportunities for the capture and use of provenance information, for example to inform reuse and retention. The use of this 'internal' provenance can inform many tasks in the wider CBR process, such as case-base maintenance, explanation, and confidence estimation. Conversational CBR and other interactive methods can involve data from user feedback on the effectiveness of proposed solutions. How to capture, represent and exploit provenance information in CBR involves many open questions. For example, the difference between the provenance recorded for data mined in case creation and the internal provenance of cases which were derived through adaptation is not clear. Furthermore, issues on the representation, storage, and maintenance of metadata and provenance within a CBR system are important to tackle in order to improve the quality of strategies exploiting this information.
The workshop’s broad goals include: (1) clarifying the nature of provenance, trust, and reputation, as they relate to CBR; (2) examining how provenance information may be used at multiple points in the CBR cycle, and (3) advancing the state of the art in relation to how provenance and meta-data should be captured, represented, and exploited in CBR systems.