Eeciently Eliciting Many Probabilities Online

In order to build complex Bayesian models from experts' input, we must elicit many probabilities; to assure continuing expert assistance , the process must be as painless and quick as possible. Previous work on the elic-itation problem has occurred in two areas: online tools and by-hand methods designed to improve accuracy and increase speed. We have merged these two streams in HYPO, the Help with Your Probabilities Online tool. This new tool combines four modes of elici-tation (table; visual; verbal; default distributions) and allows the user to navigate easily amongst the cases, and to copy, paste, and modify existing distributions. HYPO is exible, easy to use, multiuser , multi-project and scalable. Unlike elicita-tion tools that are part of speciic inference packages, HYPO is designed as a stand-alone tool, with the primary purpose of conveniently eliciting many diverse probabilities fast. It is online and available for any Bayes net building project. The input is a Bayes net structure encoded in XML, and the output is available in the XMLBIF (XML Bayesian Interchange Format) or as XML-encoded Semistructured Probabilistic Objects 2]. An ooine version also exists.