|
Scientific applications are quickly becoming more sophisticated and capable of producing breakthrough results that some scientific communities have been requesting for many years. At the same time that scientific applications are becoming more capable of responding to scientists’ requests, application results are becoming more difficult for scientists to understand and accept. The complexity of the underlying technologies supporting sophisticated applications may explain why some scientists are reluctant to accept results from complex applications. For instance, scientific workflow specifications often used in support of cyber-infrastructure applications can be instantiated by hundreds or even thousands of web services accessing and processing data gathered by large scientific communities or streamed in by swarms of large-scale sensor networks. With this not-so-futuristic context in mind, we can say that when most current complex scientific applications return results, many scientists do not know what information sources were used, when the sources were updated, how reliable the source was, or what information was looked up versus derived. Many scientists also do not know how results were derived.
This talk is centered on a single scientific question in the context of geophysics, i.e., a scientist’s request, and on the problem a scientist may face when deciding whether to accept a map as a quality result for his/her request. The acceptance of maps as quality results becomes even more challenging when the scientist acquires multiple maps, not all of them similar, yet all of them are possible results for the scientific question. We first discuss the use of provenance and provenance visualization as key mechanisms for scientists to understand maps as scientific results. In particular, we present the Inference Web approach that aims to take opaque scientific results and make the results more transparent by providing infrastructure for presenting and managing explanations. To assess our provenance solution, we present a user study that statistically demonstrates the need for scientists to use provenance information to correctly identify and explain map imperfections, if any, and thus to determine map quality.
Later in the talk, we discuss the use of trust management as a way of reinforcing scientists’ decisions concerning acceptance of maps as quality results. A comprehensive representation of trust that includes the notions of distrust, ignorance and vagueness is used to encode trust relations between agents providing spatial information used to derive map results, e.g., people, organizations, sensor networks. A topic-driven, web-based information extraction approach has been used to learn trust relations about people and organizations in the geoscience community. With the combined use of provenance and the geoscience trust network, we compute a new layer on top of maps generated by cyber-infrastructure applications and demonstrate the potential benefits of using the new layer to support a quality decisions.
| |
