Page 153 -

142 M. Adams
Chapter Notes

Worklets

The worklet approach arose from an investigation into why workflow systems had
difficulties supporting flexible processes. A strong disconnect was found between
the modeling and execution frameworks supplied, and the way human work was
actually performed. Simply put, existing systems had grown from programming or
linear, sequential execution bases, and most work activities are far from linear. These
findings were first reported in [24], which also detailed a set of criteria against which
the systems could be measured for their ability to support flexible processes. The
idea of worklets was first published in [26], and the worklet approach to dynamic
flexibility was further detailed in [27]. A full exploration of the worklet approach,
including a complete formalization and exemplary studies, can be found in [23].
For the interested reader, a discussion of the use of worklets in very creative
working environments may be found in [237].

Ripple-Down Rules

Ripple-Down Rules were first devised by Compton and Jansen [62]. While on the
surface it may seem that the quality of RDR sets would be dependent on the inser-
tion sequence of new rules, and may be open to the introduction of redundant and/or
repeated rules, this has been shown to be not the case. In terms of the correct selec-
tion of the appropriate rule based on case context, it is always the case that the
correct rule is chosen, regardless of the insertion sequence of rules into the tree. In
terms of the potential for redundancy and repetition of rules throughout the tree,
studies have shown that the issue is far less serious than first perceived [63,156] and
that the size of an RDR set which includes a normal distribution of redundant rules
compares favorably in size with other various inductively built Knowledge Based
Systems.
In terms of the number of computational steps required to reach the finally chosen
rule, it has been empirically shown that Ripple-Down Rules are able to describe
complex knowledge systems using less rules than conventional “flat” rule lists [62,
99,234]. So, when comparing insertion sequences between RDR sets and traditional
decision lists, RDR sets will have the higher quality in this regard. In terms of the
potential for degradation of computational time taken to traverse a rule set due to
the growth of rule sets over time, a number of algorithms exist for the reordering
and optimization of rule trees (cf. [99, 211, 234]). Thus, trees may occasionally be
optimized and thus the highest quality can be maintained over time.

Activity Theory

For more information on Activity Theory, the interested reader is directed to
[24,39,139].