Building Domain Models by Novices in Stochastics: Towards the Probabilistic Semantics of Verbalized Stochastic Relations

In this paper we describe a knowledge acquisition method which makes it possible to teach novices to construct Bayesian network models of their own domain. We and others had to realize that there is a severe knowledge acquisition bottleneck. It is nearly impossible to teach novices how to construct Bayesian net models of their own domain because of the huge number of conditional probabilities that are needed to describe the links of the Bayesian directed acyclic graph (dag). Because of this you have to use "toy" data from textbook examples. This leads to motivational problems because novices are often willing to adapt a new methodology only when it promises an efficiency gain in solving problems without imposing new ones. They expect at least in principle a solution sketch which feasibility can be demonstrated. So we offer the possibility that the students can describe a model of their own domain in verbal terms. The system compiles these statements into a dag. Furthermore, in this paper two methods are proposed that allow the acquisition of quantitative data from the verbally stated qualitative information. The former is needed for the application of the Bayesian network for inference tasks. One of the methods is based on likelihoods, the other one is based on frequency distributions. An important advantage of the latter method is that it substantially reduces the number of necessary knowledge acquisition steps. Both methods enable the plugin of probability tables which denotate the links of the dag, and which are a necessary part of a Bayesian net. Thus the tables are solely derived from verbal statements about stochastic relations. It is no problem to obtain these verbalizations from domain experts

[1]  Judea Pearl,et al.  Probabilistic reasoning in intelligent systems - networks of plausible inference , 1991, Morgan Kaufmann series in representation and reasoning.

[2]  P. Suppes.  A Probabilistic Theory Of Causality , 1970 .

[3]  Wibecke Brun,et al.  Yes, but it is uncertain: Direction and communicative intention of verbal probabilistic terms , 1995 .

[4]  Claus Möbus,et al.  Acquiring Knowledge from Linguistic Models in Complex, Probabilistic Domains , 1996 .

[5]  Richard E. Neapolitan,et al.  Probabilistic reasoning in expert systems - theory and algorithms , 2012 .

[6]  Max Henrion,et al.  Uncertainty in artificial intelligence: Is probability epistemologically and heuristically accurate? , 1987 .

[7]  David V. Budescu,et al.  A review of human linguistic probability processing: General principles and empirical evidence , 1995, The Knowledge Engineering Review.

[8]  H. Jungermann,et al.  Chapter 3 Causal Knowledge and the Expression of Uncertainty , 1993 .

[9]  M. Nakao,et al.  Numbers are better than words. Verbal specifications of frequency have no place in medicine. , 1983, The American journal of medicine.

[10]  L. Shulman,et al.  Medical Problem Solving: An Analysis of Clinical Reasoning , 1978 .

[11]  Karl Halvor Teigen,et al.  The language of uncertainty , 1988 .

[12]  Claus Möbus,et al.  An Intelligent Problem Solving Environment for Designing Explanation Models and for Diagnostic Reasoning in Probabilistic Domains , 1996, Intelligent Tutoring Systems.

[13]  David L. Post,et al.  Medical Problem Solving: An Analysis of Clinical Reasoning , 1979 .

[14]  David Heckerman,et al.  Probabilistic similarity networks , 1991, Networks.

[15]  Claus Möbus,et al.  Acquiring qualitative and quantitative knowledge from verbal statements and dialogues in probabilistic domains , 1997 .

[16]  Amnon Rapoport,et al.  Measuring the Vague Meanings of Probability Terms , 1986 .