Extensible Operators for Comparison Queries of Uncertain Graphs
Observational scientists study animal societies in their natural settings to understand their social relationships and behavior. Such social network data can be captured as a graph, where nodes represent observed animals and edges represent common sightings. A researcher observing a particular animal may be uncertain about its identification or behavior. This uncertainty can be expressed as existence probabilities associated with nodes and/or edges, and attribute value probabilities. Comparing these uncertain graphs is necessary to answer questions about changes in animal sociality over time, similarities of local animal communities to the general animal population, differences among animal subgroups across locations, and observation bias across researchers, to name a few. A variety of tools, databases, algorithms, and research consider analysis of probabilistic data. Similar, but mostly disjoint work has been done on graph analysis. To the best of our knowledge, there are no existing tools or research results that combine both in a general way for the purpose of uncertain graph comparison. The thesis addresses this gap by proposing a collection of operators for comparing uncertain graphs and a supporting extensible system architecture for using these operators in queries. We also present a prototype implementation of the system architecture that includes a basic query language in an existing visual graph mining tool and demonstrate the utility of the proposed operators and architecture using data from the Shark Bay Dolphin Research Project. The main contribution of this work is the development of a system architecture that facilitates creating new operators as compositions of existing ones or as new, independent ones. As demonstrated, this architecture could serve as the basis for developing simple query languages that enable scientists and others to write their own ad-hoc uncertain graph comparison queries without extensive programming knowledge.
MetadataShow full item record
Showing items related by title, author, creator and subject.