KEEPER and Protege: An Elicitation Environment for Bayesian Inference Tools
Mike Pool, Jeffrey Aikin · 2004
In this presentation we will discuss the role of Protege in a system designed for eliciting and reasoning with probabilistic models. Information Extraction and Transport, Inc. (IET) is developing the Knowledge Elicitation Environment for Probabilistic Event and Entity Relation (KEEPER) system, a tool for eliciting, storing, updating and implementing probabilistic relational models. A key feature of this tool, and focus of this presentation, is the KEEPER’s ability to elicit probabilistic relational models (PRMs or RPMs) from different sources including subject matter experts. The KEEPER elicitation component implements a single ontology for purposes of constraining and guiding elicitation and for purposes of providing the semantic bedrock for the reintegration of diverse sources and learning from diverse sources. This ontology guides and constrains the probabilistic models created by users. The KEEPER system also implements a first-ordering reasoning tool to support querying and learning and to facilitate the implementation of the PRMs in actual data scenarios. The KEEPER ontology, or tactical modeling language (TML), is stored in Protege and acts as the domain language within which all KEEPER knowledge is represented. The TML can be extended by users, but all terms used in the KEEPER knowledge base must be defined within the TML. However, in this presentation we focus on our utilization of Protege as a tool for the elicitation of PRMs and the implementation of those models in IET’s suite of uncertainty reasoning tools, Quiddity*Suite. We will address three distinct issues: • Implementing Probabilistic relational models in the