Repertory Grid Technique
The repertory grid technique is used in many fields for eliciting and analysing knowledge and for self-help and counselling purposes. The technique is essentially matrix-based although it is more complex than simply filling-in a matrix of elements. When used in knowledge engineering, the technique usually involves the following four main stages (as described below). The Rep Grid Tool in PCPACK v2 provides full support for each of these stages.
In stage 1 the concepts (called elements) are selected for the grid. For the technique to be successful and not take too much time to operate, the number chosen should be no less than about 7 and no more than about 15. A set of about the same number of attributes (called constructs) is also required. These should be such that the values can be rated on a continuous scale. The attributes can be taken from knowledge previously elicited or generated during the session using triadic elicitation.
Stage 2 involves the rating of each concept against each attribute. A numerical scale is often used, say 1 - 9. For instance, if the concepts are planets in the solar system, each might be rated on its distance from the sun (1 meaning close to the sun, 9 meaning far away), and so on through the other attributes.
In stage 3, the ratings are applied to a statistical calculation called cluster analysis to create a focus grid. These calculations ensure that concepts with similar scores are grouped together in the focus grid. Similarly, attributes that have similar scores across the concepts are grouped together in the focus grid. An example of a focus grid for planets of the solar system is shown in the figure below.
The structures to the bottom and to the right of the gird shown above are dendrograms that indicate the strength of correlations. For instance, the lower dendrogram shows Neptune and Uranus as being very similar planets, and the right-hand dendrogram indicates a correlation between size and lack of density.
In stage 4, the knowledge engineer walks the expert through the focus grid gaining feedback and prompting for knowledge concerning the groupings and correlations shown. If appropriate, extra concepts or attributes are added and then rated to provide a larger and more representative grid. In this way the technique can be used to uncover hidden correlations and causal connections.
Other Knowledge Acquisition Techniques:
Protocol Analysis techniques
Limited-Information and Constrained-Processing tasks