MICHAELA BAUMANN, M.SC. ON BEHAVIORAL PROCESS MODEL SIMILARITY MATCHING A CENTROID-BASED APPROACH MICHAELA BAUMANN, MICHAEL HEINRICH BAUMANN, STEFAN JABLONSKI THE TENTH INTERNATIONAL MULTI-CONFERENCE ON COMPUTING IN THE GLOBAL INFORMATION TECHNOLOGY (ICCGI 2015) OCTOBER 11-16, 2015 - ST. JULIANS, MALTA
2 PROCESS MODELLING Business process models used for Information system implementation (Re)design of actual workflows Documentation (QM, law) Schematical representation: C F A B E H Task Parallel split Exclusive split/ decision D G
3 PROCESS MODEL SIMILARITY MATCHING: PURPOSE AND CHALLENGES Merge multiply modelled processes Detect and reuse similar subprocesses Compare with reference process for compliance Differing vocabulary Manager A B C Differing granularity Repository A Technician B' B'' C' C''
4 OBJECTIVES Find corresponding process model elements Get a similarity value to measure similarity (between 0 and 1) Scan Read and CV Scan CV CV Correspondences? Case study and workshop Job interview Similarity value? Interview Make assessment Make overall assessment Application procedure I Application procedure II
5 PROCESS MODEL CHARACTERISTICS Correspondences not only in What has to be done? task descriptions (functional perspective) but also in involved agents (organizational perspective) non-human resources (operational perspective) data objects (data/dataflow perspective) execution modality/order (behavioral perspective) Who can do it? With what can it be done? What input/output is produced? When can it be done? Orthogonal perspectives Separate handling also for similarity determination
6 PREVIOUS METHODS AND THEIR PROBLEMS 1-to-1 correspondences granularity problem! Lexical matching well researched, but enough? Behavioral matching interleaving with lexical similarity (no orthogonality!) applicable only for 1-to-1 correspondences Scan Read and CV? Scan CV CV Case study and workshop? Job interview Make assessment Interview. The applicant is asked questions Make overall assessment
7 NEW APPROACH M-to-N correspondences granularity problem Matching for all five perspectives (among them description and behavior) separately suitable for M-to-N correspondences Scan Read and CV Scan CV CV Case study and workshop Job interview Make assessment Agent A Pen Template T1/Filled T1 Interview. The applicant is asked questions Make overall assessment
8 SIMILARITY FOR PROCESS PERSPECTIVES UNDER M-TO-N CORRESPONDENCES Functional perspective: Concatenation of descriptions String operations, stemming, ontologies, etc. CV Scan and read CV Scan Scan CV Read and CV Organizational, Operational, Data/Dataflow perspective: Set-based methods on subject/object identifiers Agent Set 1 Agent Set 2 What about behavioral perspective?
9 BEHAVIORAL PROCESS MODEL SIMILARITY Idea: Split up behavior into three orthogonal dimensions assignable to each task position π repeatability ρ optionality ο Determine a centroid for each dimension for each set of tasks induced by the M-to-N mapping Determine similarity through averaged 1-minus-distance of the centroids CV Job interview Make assessment π: 0.2 ο: 0 π: 0.4 ο: 0 π: 0.6 ρ: 1 ο: 0 π: 0.8 ο: 0
10 BEHAVIORAL PROCESS MODEL SIMILARITY GRAPHICAL APPROXIMATION Given: M-to-N correspondences (bijective mapping) indicated through colors CV π: 0.2 + 0.4 2 = 0.3 π: 0.07 1 6 + 2 6 + 1 5 = 7 3 30 0.23 Scan Scan CV Case study and workshop Read and CV Job interview π: 0.6 π: 0.6 0 Interview Make assessment π: 0.8 π: 0.8 0 Make overall assessment positional similarity: 0.98
11 BEHAVIORAL PROCESS MODEL SIMILARITY GRAPHICAL APPROXIMATION Given: M-to-N correspondences (bijective mapping) indicated through colors CV + 0 2 = 0 0 Scan Scan CV Case study and workshop Read and CV Job interview ρ: 1 1 Interview Make assessment 0 Make overall assessment repeatability similarity: 0.67 optionality similarity: 1
12 VALIDATION Comparison of three methods for measuring behavioral similarity Causal footprints Smallest causal footprints Centroid-based approach A Models G 1 and G 2 show a more similar behavior than models G 1 and G 3 G 1 G 2 B C D AB C DE E G 1 -G 2 G 1 -G 3 CF 80% (294) 64% (414) SCF 88% (90) 63% (108) CBA 100% (30) 33% (30) G 3 A E C B D Similarity and number of calculated intermediate values
13 CONCLUSION Starting point: Need for determining similarity of process models Different process perspectives (availability and quality) Various similarity measuring methods for 1-to-1 correspondences Contribution: Similarity measure for behavioral perspective separated from other perspectives and suitable for M-to-N correspondences Result: Assignment of behavior to tasks via three dimensions (position, repeatability, optionality) Centroid-based behavioral similarity measure Low computational effort Still missing (but in progress): Implementation