As far as I have understood, Bayesialab supports basically three methods to elicit expert knowledge, when a consensus about DAG itself is available: with Bekee, by using equations to define the dependences between the variables and by filling the conditional probability tables for each variable manually. One could say, that the situation is very good. Compared to many other tools, the situation is indeed good. But for certain type of use cases, which might be the biggest opportunity, the situation could be much better:I have realized this in certain research papers before, and again now in two recent studies, which cover risk mangement by using Bayesian Networks (the one handles tolerance of Public Safety and Security Networks against certain type of risks, to be published soon, and another assessing information security risks of mobile phones, a Master Thesis, where I have been advisor). The problem in both has been, that the experts are very busy and often don't want to learn some new concepts or tools (such as Bekee) and have not time nor motivation to to fill the CPTs manually. Also sometimes a reasonable good mathematical background is lacking in order to formulate the dependency parametrically.I assume, that the majority of expert knowledge elicitation use cases and situations are such, that someone, who can use the proper tool such as Bayesialab, should "interview the experts", one at a time, and the output from this interview session is a Bayesian Network as a document from his/her knowledge. The same for each expert knowledge. The requirement of course then is, that there must be some fast and innovative ways to elicit within one or two hours the knowledge of experts by the interviewing and by using Bayesialab tool. Bekee is doing this but in it idea has been, that the expert is putting effort to learn and use the tool.I am proposing to Bayesia guys to think the following, when planning future releases of Bayesialab: a) How and what are the simple steps to elicit a DAG (qualitative network) based on and during one hour interview session with a expert?b) How to aggregate the achieved DAGs (each of them represents knowledge of one expert) to one DAG which describe a consensus from each expert? This kind of efort will be done by interviewee and by using the tool.c) How and what are the simple steps to elicit the CPTs for each variable in DAG (quantitative network) based on and during one hour interview session with n expert by using the concensus DAG as a basis?d) And finally how to aggregate these DAGs with CPTs to represent a consensus about a quantitative DAG.A sort of "personal Bekee" might be a right approach for the above?
Quote 0 0