Presentation of the paper "Discovering Unbounded Synchronization Conditions in Artifact-Centric Process Models" at the 9th International Workshop on Business Process Intelligence (BPI'2013) - http://www.win.tue.nl/bpi2013/

1. Viara Popova and Marlon Dumas
University of Tartu
Discovering Synchronization
Conditions in Artifact-Centric
Process Models

2. Artifact Model Discovery Tool
Raw Log
(text or DB)
Discover
Artifact Types &
Associations
Extract Artifact
Logs
Discover GSM
per Artifact
Discover Intra-
Artifact
Conditions
Discover
Synchronization
Conditions
GSM Artifact
Model
BizArtifact
(Barcelona)
2

3. Raw Logs
2008-12-09T08:20:01.527+01:00 Received_order 245BG „Metallica” „Dead Magnetic”
2008-12-09T20:11:15.342+01:00 Received_order 246BL „Ray Baretto” „Acid”
2008-12-10T08:22:01.427+01:00 Sent_quote 245BG
2008-12-10T08:30:01.427+01:00 Sent_quote 246BL
2008-12-11T11:20:14.534+01:00 Accepted_quote 246BL
...
Timestamp
Event type
Data attribute
Data attribute Data
attribute
3

4. Raw Logs  Artifact Logs
1. Discover entities
 Functional dependencies  primary keys
2. Discover relationships
 Inclusion dependencies  foreign keys, multiplicities
3. Discover artifact types
4. Extract logs of each artifact
4
Material Order
(MO)
Customer PO
Artifact log
Artifact
logs
*1

5. Discovered GSM
Material Order (MO)
t1: Create
MO
on create t1
complete
t2: Send MO
to supplier
on t1
complete
t2
complete
t3: Receive
suppl. response
on t2
complete
t3
complete
t4: Receive
items
on t3 complete
on t4
complete
t5: Receive
invoice
on t3
complete
t5
complet
e
t6: Reassign
supplier
on t3
complete
t6
complete
t7: Close
MO
on t4 complete
and t5 complete
t7
complete
5
One artifact, pure control-flow

6. GSM Discovery Tool Chain
Raw Log
(text or DB)
Discover
Artifact Types &
Associations
Extract Artifact
Logs
Discover GSM
per Artifact
Discover Intra-
Artifact
Conditions
Discover
Synchronization
Conditions
GSM Artifact
Model
6

7. Composition of guard sentries
A guard sentry includes:
1. Intra-artifact data conditions
 “weight > 100”, “response = positive”
 Discovered using ProM decision
miner or BranchMiner
2. Inter-artifact synch conditions
7

8. GSM with intra-artifact conditions
(MO)
t1: Create
MO
on create t1
complete
t2: Send MO
to supplier
on t1
complete
t2
complete
t3: Receive
suppl. response
on t2
complete
t3
complete
t4: Receive
items
on t3 complete &
positive response on t4
complete
t5: Receive
invoiceon t3 complete
& positive response
t5
complet
e
t6: Reassign
supplier
on t3 complete &
negative response t6
complete
t7: Close
MO
on t4 complete
and t5 complete
t7
complete
8

9. Inter-artifact synch conditions
Points of synchronization between artifacts:
the transition to a new state of one artifact is
triggered by the states of related instances of
another artifact
Paper can only be evaluated when at least three
reviews are completed.
Meeting can only be confirmed when at least half
of the members have confirmed participation.
Akin to completion conditions in BPMN multi-
instance activities
9

10. Synch condition discovery
10
Primary
artifact
Secondary
artifact

11. Synch condition discovery (cont.)
11

12. Synch condition discovery
12
Primary
artifact
Secondary
artifact

13. Find synchronization points
Heuristic to find probable points:
Activity level of S – average number of events
in secondary artifact happening immediately
before S
A B C D E S F G D S
2 1
The higher the activity level the more likely S is a
synchronization point
13

14. Find conditions for point S
 For each execution of S: a snapshot of the current
states in the related instances
 Feature vector:
 For each event type T in secondary artifact: how
many instances were in state T (T last executed)
when S was executed
 Positive examples: one vector per execution of S
 Negative examples: one vector per execution of
other event in main artifact + one vector per
execution of an event in secondary artifact
A B C D E S F G D S
S: positive examples: (B:0,D:1,E:1)(B:0,D:1,E:0)
14

15. Find conditions for point S
Refinements:
Remove redundant samples
Balance number of positive and negative
examples
Decision tree  synch conditions
Scoring each condition:
Quality of decision tree – F-measure
Size of decision tree (normalized)
Activity level (normalized)
15

16. GSM with synch conditions (PO)
t1: Record
PO
on create t1
complete
t2: Analyze
PO
on t1
complete
t2
complete
t3: Generate
MOs
on t2
complete
t3
complete
t4: Assemble
Product
on t3 complete &
all MOs fulfilled on t4
complete
t5: Receive
payment
on t3
complete
t5
complet
e
t6: Notify
customer
on t3 complete
& MOs
unfulfilled
t6
complete
t7: Close
PO
on t4 complete
and t5 complete
t7
complete
16

17. ArtifactMiner – Implementation
discovery
(reverse-
engineering)
Artifact model
repair
conformance /
A
B
A
C
A
B
Artifact logs
raw event log
DB
Artifact
Types
Guards
(OCL)
Artifact
Lifecycle (GSM)
17

18. Evaluation
Synthetic logs (x 2 sample processes)
All sync conditions found
FRIS log
 Flemish applied research funding agency
 TBM Funding Program (biomedical research grants)
 Extracted from existing Sharepoint system (one
funding call, 2 artifact types, 200+ events)
 Correctly discovered artifacts + control-flow
 Discovered 4 relevant synchronization conditions
 No missing condition (given the available data)
18

19. Looking Ahead
Exciting case studies
TU/e + KPMG
Extracting artifact models from ERP systems
(e.g. SAP, Oracle)
Discovering BPMN models with multi-instance
activities & synchronization conditions
19

20. QUESTIONS?
Research funded by EU’s FP7 Programme (ACSI Project)
20