Information Technology Reference
In-Depth Information
arrive
p
miss
, then we consider the
FlowData
as satisfied and activate the
corresponding arc.
Let is consider the incoming arc of
Billing and Invoicing
[G] from the scenario:
messages (5b) and (6) always occur thus their probability is 1. Assume (5b)
arrives then rule 2 applies, as we are still waiting for (6). Once message (6)
arrives, rule 3 would consider the arc conditions satisfied and subsequently enable
activity [G].
When the observed messages suce to activate the underlying arc, we it-
erate through all non-observed alternative messages (that still reside in state
Scheduled
,
Expected
,or
M issing
) and switch them to
NotExpected
as we no
longer require their appearance.
5.2 Message Prediction
Message prediction is purely based on the transition labels between message
states. In any of the
W aiting
states the probability is determined based on the
transition to the respective received state. From any of the
Arrived
states the
probability is determined by the transition weight to the
Repeated
state.
After any change to any of the messages' state model, the messages are ranked
according to their probability to occur in their current state. The top ranked
message(s) constitute the prediction and are applied during the classification
of newly intercepted messages. Suppose we have two messages and their corre-
sponding message state models depicted in Figure 5(a) and (b): the first one
currently in state
Expected
, the second in state
M issing
. Hence, we would pre-
dict the occurrence of message (b) with
p
occur
(
b
)=0
.
9 rather than message (a)
with
p
occur
(
a
)=0
.
7. We are thus able to address and manage messages that
over multiple process instances no longer adhere to the process model but have
become delayed.
Received
Early
Received
Early
Received
Early
Scheduled
Scheduled
Scheduled
1
1
1
0.7
0.2
0.14
Received
On Time
Received
On Time
Received
On Time
0.1
Expected
Expected
Expected
0.8
0.86
0.3
Repeated
Repeated
Repeated
0.9
0.63
Received
Late
Received
Late
Received
Late
Missing
Missing
Missing
1
1
0.9
Not
Expected
Received
Unexpected
Not
Expected
Received
Unexpected
Not
Expected
Received
Unexpected
1
1
1
1
0.1
0.37
Never Occurred
Occurred
Never Occurred
Occurred
Never Occurred
Occurred
(a)
(b)
(c)
Fig. 5.
Examples of message state model instances: only non-zero transitions are in-
cluded for sake of clarity. Subfigure (a) displays a message that when expected occurs
with 70% probability and is repeated in 10% of process cases. Subfigure (b) describes a
message that occurs only 20% on time, but still arrives in 90% of all cases when miss-
ing. The effect on the transition probabilities of that message never occurring (thick
lines and labels) in a single process instance is depicted in subfigure (c).
