This presentation describes the new BayesiaLab Knowledge Elicitation Environment. This environment allows reducing biases (cognitive, group and facilitator), and allows to greatly improve the traceability of the brainstorming session.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Every Company is faced to complex decisions that need to be rationally supported
Sometime, there are too few data available, or no data at all, to allow using data mining and data analysis technics to automatically build a Decision Support System
Experts have gathered invaluable Tacit Knowledge through their experience
We need to Convert this Tacit Knowledge into Explicit Knowledge and use it to build a model
We want actionable models to allow What-if scenarios (simulation and/or diagnosis), drivers analysis, ...
Bayesian Belief Networks (BBNs) are ideal models for such problematics: their graphical representation allows a manual design by using expert knowledge, and their probabilistic engines offer powerful simulation capabilities
Clear definition of the BBN’s objective(s) (e.g.: Improvement of the Product/Service Quality, improvement of the Purchase Intent, improvement of the Company’s performance, ...)
Identification of the conceptual dimensions that are linked to those objectives (e.g.: Human resources, Management, Production, Marketing, ...)
Definition of the group of experts that will fully cover all the dimensions (and the different geographical zones), with a small redundancy to allow fruitful debates
Brain Storming Sessions with this group of Experts to manually build the BBN
Definition of the exclusive states of those variables
Creation of one node per identified variable
Brainstorming to define the direct relationships between the variables, and addition of the corresponding arcs between those dependent variables
The structure elicitation is probably the simplest task of the Brainstorming session
For each root node, i.e. without incoming arc, definition of the marginal probability distribution over the defined states
For each node with incoming arc(s), definition of the conditional probability distribution over the defined states, for each combination of the states of its connected nodes
Each expert gives his/her belief on the distributions
There are various kinds of biases to be aware of
Cognitive (Plausibility, Control, Availability, Anchoring) Emotional (Mood, Motivation) Group (Anchoring, Herding) Facilitator (can be biased toward charismatic experts or toward
This Expert Editor allows defining:The Expert’s name, its Credibility (that will be use globally during the consensus computation), her/his Picture, a Comment to describe her/his area of expertise. The last field contains the number of assessments realized by the expert on the
Selecting a cell in the probability table activates the Assessment button for assessing the question corresponding to the selected line, i.e. what is the marginal probability
Once the Expert validates her/his assessment, this assessment is sent to the BayesiaLab’s server and the Facilitator’s listener is automatically updated
This listener allows following the status of the
Experts’ assessments
Clicking on OK makes BayesiaLab harvesting the
assessments. Closing the window cleared the question from the webpage of the Experts that do not have
A small icon is added at the left of each probability to graphically
represent the consensus degree:from a full transparency when there
all the Experts agree on the probability, to no transparency when the range of the assessments is 1
Once the assessments validated, a Mathematical consensus is computed by using the Experts’ credibility and their assessment’s confidence. This automatic consensus can be manually
modified by the Facilitator to set a Behavioral consensus, i.e. one issued after a fruitful debate
Hovering over this icon returns the minimum and the
maximum assessments, and the number of assessments
Pressing the “i” key while hovering over the expert icon allows displaying
the information panel below
This information panel contains:- the number of rows ((Conditional) probability distributions) that comes with Experts
assessments- the total number of assessments that have been set in the probability table
- the number of Experts that have assessed at least one probability distribution in the table - a measure of the global disagreement that takes into account the deviations from the
mathematical consensus- the maximum disagreement corresponding to the greatest difference between two
This exportation tool generates a CSV file with all the assessments of the Experts.
There is one column per Expert, one column per Expert’s Confidence (yellow), the last column indicating the weight of the line (1/number of states of the assessed variable)
(green).Each line describes the Experts’ assessment
Each color corresponds to a cluster. Three segments of Experts have
been induced in that example. The real experts behind those anonymized
experts have indeed three different profiles (functionally and geographically)
Dendrogram corresponding to that
segmentation
Based on the obtained Expert Segments, one Bayesian network per segment can be generated (by using the Expert Editor). This can be useful for analyzing the sensibility of the model, but also to get specific networks (depending on the geographical localization
for example)
28
Based on the obtained network, Experts can be clustered into homogeneous groups by using the BayesiaLab’s Variable Clustering algorithm
Three kinds of analysis are available, depending on the Random selection policy that is chosen to generate the set of networks (1000 networks in the above example):
- One Expert per network: each network generated is parametrized by using the selected Expert (or the consensual probability if the selected Expert has not been involved in the
assessment)- One Expert per node: each network generated is parametrized by selecting for each node one Expert. If the selected Expert is not involved, the consensual probability if the selected
- One assessment per Conditional Probability Table’s row (if any)