Information Technology Reference
In-Depth Information
Probabilistic Process Engine
2
Message State
Model Tuning
3
Process
Progress
Update
Message
State
Update
7
Process Model
Adaptation
Msg &
Action
Correlation
Msg &
Activity
Ranking
6
Process Model
& Msg State
Model
Sequence
Analysis
1
4
5
Messages & User
Actions
Message & Activity
Recommendation
Fig. 2.
Supporting message prediction and activity recommendation through self-
learning message flows
advance the progress of the process (2). Any completed activity potentially re-
sults in an update of one or more messages, respectively their expectation states:
messages are activated, become missing, or should no longer occur (3). Finally,
the engine ranks messages according to their probability to occur, and activities
to be carried out next (4). Activity recommendations support the user in car-
rying out his/her work. Message recommendations support the classification of
unstructured and semi-structured messages that are exchanged between process
participants.
After successful completion of a process instance, the mechanism we present in
the following sections analyzes the sequence and timing of activities and messages
(5) to update the process model (6) and message state model (7) to accurately
reflect the changes in the real world.
Our approach specifically targets dynamic business environments that mostly
rely on unstructured or semi-structured messaging to communicate and co-
ordinate processes. SMEs and event organization usually coordinate and col-
laborate via email. Our approach relies on an infrastructure for intercepting
those messages, extracting relevant information, and mapping those messages
to document types. In our case, the EU FP7 Project Commius provides the
necessary framework to extract information from emails and conduct a basic,
process-unaware email content analysis. Details on the actual process orches-
tration as well as message interception, extraction, analysis, and user inter-
faces are outside the scope of this paper. The interested reader is referred to
previous project-related publications [4,13,11]. Our prototype implements the
algorithms and techniques introduced in this paper to support the email con-
tent analysis by determining the expected document types. The activity rec-
ommendations allow the annotation of emails with process-relevant information
(as defined in the activity description) before forwarding them to the actual
recipient.
