|Institution:||University of Illinois – Urbana-Champaign|
|Full text PDF:||http://hdl.handle.net/2142/24020|
Communications through phone, email and social networking create increasingly vast webs of data that can be analyzed for research in areas such as marketing, forecasting, and criminal network analyses. Trained analysts can manipulate algorithmically processed data to find conclusions for further investigation, and visualization of this data can enable analysts to draw conclusions more quickly and accurately than from text alone. To examine various methods for manipulating visualized, networked data an experiment was conducted in which participants used an interactive program to determine if hierarchical sub-graph hypotheses conformed to network data. To determine what types of training best prepare analysts to interact with such networked data sets, participants were divided into four groups, each completing one of four training exercises prior to the interactive visualization portion of the experiment. Training sessions required participants to examine or actively create visual, verbal, or visual and verbal hierarchical structures. Groups that were trained with multiple representations of hierarchical structures correctly completed more hypothesis testing tasks and were able to do so more quickly. All groups showed an overall reduction in time needed to correctly complete the hypothesis testing tasks as they proceeded through trials, resulting in learning curves for each group. Various solution methods were identified that were used by high-performing subjects, but differences in method suggest that analysts create individualized mental-models, and there may be no single ???best??? solution method.