Mixed-initiative data mining with Bayesian networks
Robert F. Stark, Michael P. Farry, Jonathan Pfautz · 2012
Complex information systems make a wide variety of data types available, but users may find it difficult to obtain insight when inspecting those data sets. This need has led to data analytics research and resulting technologies such as information visualization and data mining. While these research efforts provide a necessary and useful component of many information systems, they lack the ability to capitalize on both human and computer capabilities. A mixed-initiative approach to data mining would enable the integration of human and machine capabilities for search and review of data. Because Bayesian networks (BNs) allow for deductive and abductive reasoning under uncertainty, they are a good fit for supporting human-computer collaborative data mining. To support mixed-initiative data mining that capitalizes on BN strengths, we present a technical concept and task flow in which the human and computer work collaboratively to construct a joint knowledge model from a complex data set. We see this task flow being useful for knowledge acquisition, situation assessment, and business intelligence.