Slides of the paper presented at the 34th International Conference on Advanced Information Systems Engineering (CAiSE 2022, Leuven, Belgium)
Abstract:
Process mining techniques provide process analysts with insights into interesting patterns of a business process. Current techniques have focused by and large on the explanation of behavior, partially by help of features that relate to multiple perspectives beyond just pure control flow. However, techniques to provide insights into the connection between data elements of related events have been missing so far. Such connections are relevant for several analysis tasks such as event correlation, resource allocation, or log partitioning.
In this paper, we propose a multi-perspective mining technique for discovering data connections. More specifically, we adapt concepts from association rule mining to extract connections between a sequence of events and behavioral attributes of related data objects and contextual features.
Our technique was evaluated using real-world events supporting the usefulness of the mined association rules.
Paper: Bayomie, D., Revoredo, K., Mendling, J. (2022). Multi-perspective Process Analysis: Mining the Association Between Control Flow and Data Objects. In: Franch, X., Poels, G., Gailly, F., Snoeck, M. (eds) Advanced Information Systems Engineering. CAiSE 2022. Lecture Notes in Computer Science, vol 13295. Springer, Cham. https://doi.org/10.1007/978-3-031-07472-1_5
3. PAGE 3
Introduction
RQ: What is the relation between control-
flow perspective and data objects behavior
perspective?
We propose a multi-perspective mining
technique based on association rule mining
to discover the data connections between
the control-flow and data objects behavior.
-
-
-
-
5. PAGE 5
Preparing the Event log
• Prepare the data to align with the process analysts analysis
objectives
Log partition-
time frame
-
-
-
-
-
-
-
-
6. -
-
PAGE 6
Encoding event log into transaction table
• A transaction table sustains the control-flow perspective and
the change behavior over the data perspective.
• A transaction represents the
behavior of the data attributes
over a pair of events.
Control-flow perspective Atomic perspective Complex perspective
12. PAGE 12
Analysing the rules - Ranking
We rank the rules using the known measures of association rules.
𝑠𝑢𝑝𝑝𝑜𝑟𝑡 𝑅 =
|(𝑎𝑛𝑡𝑒𝑐𝑒𝑑𝑒𝑛𝑡 𝑐𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑡) ⊆ 𝑇 |
|𝑇|
where 𝑇 𝑖𝑠 𝑡ℎ𝑒 𝑡𝑟𝑎𝑛𝑠𝑎𝑐𝑡𝑖𝑜𝑛𝑠
𝑐𝑜𝑛𝑓𝑖𝑑𝑒𝑛𝑐𝑒 𝑅 =
𝑠𝑢𝑝𝑝𝑜𝑟𝑡(𝑅)
𝑠𝑢𝑝𝑝𝑜𝑟𝑡(𝑎𝑛𝑡𝑒𝑐𝑒𝑑𝑒𝑛𝑡)
𝑙𝑖𝑓𝑡 𝑅 =
𝑠𝑢𝑝𝑝𝑜𝑟𝑡(𝑅)
𝑠𝑢𝑝𝑝𝑜𝑟𝑡 𝑎𝑛𝑡𝑒𝑐𝑒𝑑𝑒𝑛𝑡 ∗ 𝑠𝑢𝑝𝑝𝑜𝑟𝑡(𝑐𝑜𝑛𝑠𝑒𝑞𝑢𝑒𝑛𝑡)
13. In this step, We prepare the rules to align with the analysis objectives.
PAGE 13
Analysing the rules - combining
14. EL-RM aggregates the rules based on common antecedent and common
consequent. We propose three possible aggregations.
1. The first option focuses on the control-flow perspective.
2. The second option focuses on the data object perspective.
3. The third option focuses on combine both the perspectives.
PAGE 14
Analysing the rules - combining
15. PAGE 15
Analysing the rules - compare
• we compare the rules generated from the different partitions to detect
the changes in behaviour over the divisions.
• We induce various sets of rules from the extracted rules by using the set
operations:
All rules (Union set operation)
Common rules (intersection set operation)
Difference rules in a partition against Common rules (difference set operation)
Partition unique rules (difference set operation)
16. We conducted three exploratory experiments on three real datasets.
PAGE 16
Evaluation
BPIC-2017 contains the events of the loan application
process of a Dutch financial institute. The events are
generated from three sub-processes, i.e., application,
offer and workflow.
The log contains cases that started at the beginning
of 2016 until the 1st of February 2017.
17. PAGE 17
Experiment 1 – BPIC2017
Discovered rules = 751
Combine rules = 15
Confidence = [0.94,1]
Lift = [0.95, 13.63]
• R1: IF ei.Act = “OCreate Offer” and
ej.Act = “ OCreated”
THEN ei.EventID = ej.OfferID
• R2: IF ei.Act =”A_Complete” and
ej.Act = “ W_ValideApplication”
THEN ei.Resource = ej.Resource
18. We conducted three exploratory experiments on three real datasets.
PAGE 18
Evaluation
BPIC-2020 (prepaid travel) contains the events of
the prepaid travel request process at Eindhoven
University of Technology (TU/e).
The log covers the cases from the beginning of
2017 till the 21st of February 2019.
19. PAGE 19
Experiment 2 – BPIC2020(prepaid travel)
All rules = 425
Common rules = 85
Confidence = [0.99, 1]
Lift = [1, 8.90]
• R1: IF ei.Act =“Permit APPROVED by
ADMINISTRATION” and ej.Act =
“Permit APPROVED by BUDGET
OWNER” THEN ei.Resource =
ej.Resource
• R2: IF ei.Act = “Permit APPROVED by
BUDGET OWNER” and ej.Act = “Permit
FINAL_APPROVED by SUPERVISOR”
THEN ei.org:role != ei.org:role
20. We conducted three exploratory experiments on three real datasets.
PAGE 20
Evaluation
Road traffic fine management process dataset
contains the events of the road traffic fines process.
The cases have a diverse cycle time duration
behavior. [mention duration of shortest and longest]
The log covers the cases from the beginning of 2000
till the 18th of June 2013
21. PAGE 21
Experiment 3 – Road traffic fine
All rules = 239
Common rules = 159
Average confidence = 0.96
Average lift = 1.98
• R1: IF ei.Act =“Insert Fine Notification”
and ej.Act = “Insert Date Appeal to
Prefecture” THEN ei.NotifcationType !=
ej. NotifcationType
• R2: IF ei.Act =“Create Fine” and ej.Act =
“payment” THEN ei.NotifcationType =
ej. NotifcationType
22. We proposed a multi-perspective mining technique for the discovery of data
connection.
Our method uses association rules to represent the relation between the control
flow perspective and its impacts on the behavior of the data objects perspective
The results of our evaluation showed the potential of the approach to extract
relevant insights about the change behavior of the attributes over
the events.
PAGE 22
Conclusion