Discovering Declare Maps
R.P. Jagadeesh Chandra Bose (JC)
Joint Work with
Fabrizio M. Maggi and Wil M.P. van der Aalst
The Apriori Approach
• Discover Frequent Activity Sets (Candidate Sets)• {A, B}, {C, E}, {A, E}, …
• Generate Dispositions• (A, B), (B, A), (C, E), (E, C), (A, E), (E, A), …
• Instantiate Constraints• response (A,B), response (B,A), …
• Assess Significance and Prune Constraints• Support, confidence, interest factor, …
F.M. Maggi, R.P.J.C. Bose and W.M.P. van der Aalst. Efficient Discovery of Understandable Declarative Process Models from Event Logs, CAiSE 2012, pp 270-285
Dealing with Redundancy
Retain the strongest
F.M. Maggi, R.P.J.C. Bose and W.M.P. van der Aalst. A Knowledge-Based Integrated Approach for Discovering and Repairing Declare Maps, CAiSE 2013 (to appear)
Dealing with Redundancy
transitive reduction
Case, M.L.: Online Algorithms To Maintain A Transitive Reduction. In: Department of EECS, University of California, Berkeley, CS 294-8 (2006)
Apriori Declare Map
• Reference set of templates/activities• Repair the map
• add stronger constraints• remove constraints that no longer hold
• Use for selecting pruning metric thresholds
Extending with Data
• Issues• Too many constraints (not all may be interesting)• ambiguities in associating events
− <a, b, c, b>, <a, b, a, b>• Lack of diagnostic information
R.P.J.C. Bose, F.M. Maggi and W.M.P. van der Aalst. Enhancing Declare Maps Based on Event Correlations, BPM 2013 (to appear)
Discovering Correlations
• Relationship between attributes• continuous (<, <=, >, >=, =, !=)• string/boolean (=, !=)• timestamps (before, after, time diff)
• Comparable attributes• apriori knowledge• attributes of the same type
Framework
<a, c, b>, <a, a, c, b> (non-ambiguous)<a, b, b>, <a, a, b, b> (ambiguous)
# instances where correlation is true
# instancesSupport (correlation) =
Discovered Correlations (Example)
A = First outpatient consultation,B = administrative fee - the first polC = unconjugated bilirubin D = bilirubin- totalE = rhesus factor d - Centrifuge methodF = red cell antibody screening
Discriminatory Patterns
• Constraint activations can be classified into different categories• conformant vs. non-conformant• slow, medium, fast based on their response times• …
Framework
• Class Labeling• Feature Extraction
• feasible correlations• antecedent activity attributes• case-level attributes
• Discover Patterns
Discriminatory Patterns (Example)
• response (A,B): 517 non-ambiguous instances: 60 violations
A = First outpatient consultation,B = administrative fee - the first pol
A.Section is Section 5 AND DiagnosisCodeSet is {106; 823} then violation (TP=5, FP=1)A.Section is not equal to Section 5 AND A.Producercode is SGSX then violation (TP=3, FP=1)