Page 275 -

9.5 Recommend 257

Fig. 9.12 Recommendations
can be based on predictions.
For every possible choice,
simply predict the
performance indicator of
interest. Then, recommend
the best one(s)

best given the goal selected?”. However, the decision space may also consist of a set
of resources and the goal is then to recommend the best resource to execute a given
activity. For example, the operational support system could recommend allocating
activity h to Mike to minimize the flow time. This example shows that recommenda-
tions are not limited to control-flow and can also refer to other perspectives. There-
fore, we use the term “action” rather than activity. The decision space for a running
case may be part of the message sent from the enterprise information system to the
operational support system. Otherwise, the recommendation model should be able
to derive the decision space based on the partial trace.
As shown in Fig. 9.12, recommending an action to achieve a goal is closely re-
lated to predicting the corresponding performance indicator. Suppose that for a case
having a partial trace σ p we need to recommend some action from a set of possi-
ble actions {a 1 ,a 2 ,...,a k }. The existing partial trace can be extended by assuming
that action a 1 is selected (although it did not happen yet). σ 1 is the resulting ex-
tended partial trace, i.e., σ 1 = σ p ⊕ a 1 . (Here we assume that a 1 is an activity and
we use simple traces.) The same can be done for all other actions resulting in a set of
partial traces D ={σ 1 ,σ 2 ,...,σ k }. Now a prediction is made for the selected per-
formance indicator and each element of D. The resulting predictions are compared
and ranked. If σ 2 has the best predicted value (e.g., shortest remaining flow time),
then a 2 is recommended first.
Depending on the prediction technique used, the recommendation can also in-
clude information about its reliability/quality, e.g., the confidence or certainty that a
particular selection is optimal with respect to the goal. For example, in Fig. 9.11 the
recommendation attaches a confidence to each of the three possible actions. How to
interpret such confidence values depends on the underlying prediction method. For
example, if short-term simulation is used, then the 85% certainty of x mentioned in
Fig. 9.11 (i.e., the confidence attached to recommendation x) would mean that in
85% of the simulation experiments action x resulted in the shortest remaining flow
time.