---

Action Patterns in Business Process Models

Description

Business process management experiences a large uptake by
the industry, and process models play an important role in the analysis and
improvement of processes. While an increasing number of sta� becomes
involved in actual modeling practice, it is crucial to assure model quality
and homogeneity along with providing suitable aids for creating models.
In this paper we consider the problem of o�ering recommendations to the
user during the act of modeling. Our key contribution is a concept for
de�ning and identifying action patterns – chunks of actions often appearing
together in business processes. In particular, we specify action patterns
and demonstrate how they can be identi�ed from existing process model
repositories using association rule mining techniques. Action patterns
can then be used to suggest additional actions for a process model. Our
approach is challenged by applying it to the collection of process models
from the SAP Reference Model.

Transcript

Action Patterns in Business Process Models Sergey Smirnov 1 , Matthias Weidlich 1 , Jan Mendling 2 , and Mathias Weske 1 1 Hasso Plattner Institute, Potsdam, Germany 2 Humboldt-Universit ä t zu Berlin, Berlin, Germany Agenda

• Motivation

• Preliminary Concepts

• Co-occurrence action patterns

• Behavioral action patterns

• Evaluation with SAP Reference Model

• Conclusion

Motivation (I)

• In enterprises the modelers are of different skills and backgrounds.

• An enterprise often maintains a large repository of process models.

• An important problem is to assure the high quality of process models.

Motivation (II) ? One aspect of process model quality is if the model is complete from a business point of view. An example question is if all actions typically appearing in this business process present in the current model. How to suggest the modeler which model elements are missing in the designed model from a perspective of business semantic? Idea!

• Let us derive information for user support from existing models …

• … and capture it in action patterns – groups of actions which often appear together in business processes.

Actions in Process Models

• Observation

• Labels of model element capture the process business semantics.

• Challenge

• Several labels capture similar meaning

• Solution

• Derive actions from activity labels

• create

Modeler Assistance allocate calculate revaluate … {allocate calculate} -> {settle} {create update} -> {archive} … {allocate calculate} -> {settle} Assumptions

• Process model collection is large enough

• Activity label signifies an action

• A mechanism interpreting an activity label as an action exists

Auxiliary Concepts

• Alphabet of labels

Alphabet of terms creation of order send notification book hotel verify customer update report receive order create process validate client order hotel invoice receive send notification approve update close Process Model

• Process model is a tuple

• A is a finite nonempty set of activities

• G is a finite set of gateways

• F is a flow relation

• s is the only start activity

• e is the only end activity

Action Function

• Action function derives an action from a label

creation of order -> create send notification -> send verify customer -> verify update report -> update Process Model Collection

• Process model collection is a tuple

• is a nonempty finite set of process models

• is the set of all actions in the model collection

Association Rule Learning

• a set of items

• a transaction,

• a collection of transactions

• set of items

• satisfies , if

• Support

13 10 17 supp( ) = 40 supp( ) = 30 Association Rule

• An association rule in a collection is an implication of the form , where and .

• Confidence

conf( ) = 0.75 Co-occurrence Action Patterns

• An action is an item

• A process model is a transaction

• A process model collection is a transaction collection

Co-occurrence Action Pattern

• is a co-occurrence action pattern in process model collection , where:

• is an association rule , where

• minsup is the required minimal support

• minconf is the required minimal confidence

Co-occurrence action patterns settle allocate calculate process archive enter process permit evaluate level process pick consequent antecedent Weak Order Relation Activities a and b are in weak order relation, if there is a trace in which b occurs after a . We denote weak order relation as Example: Behavioral Profile

• 1. Strict order (A D)

• Reverse strict order (D A)

• 2. Exclusiveness (F + G)

• 3. Observation concurrency (C || E)

• Behavioral profile is the set of the 3 relations.

1 2 3 -1 Behavioral Action Patterns

• A behavioral relation is an item

• A process model containing a co-occurrence action

• pattern is a transaction

• Process models containing a co-occurrence action

• pattern is a transaction collection

Behavioral Action Pattern

• is a behavioral action pattern in process model collection , where:

• is a rule , where

• and action pairs for which behavioral relations

• are specified

• minsup is the required minimal support

• minconf is the required minimal confidence

Behavioral action pattern {allocate, calculate, settle} allocate settle, calculate settle allocate calculate consequent antecedent Approach Evaluation

• Goals

• I. check if it is possible to derive action patterns from a collection of process models

• II. learn which support and confidence values are encountered in practice

• Means

• Model Collection: SAP Reference Model

SAP Reference Model

• … captures business processes supported by SAP R/3

• … captures 29 functional branches of an enterprise

• … contains 604 Event-driven process chains (EPCs)

• … model element labels are in English

Co-occurrence pattern number minconf minsup 0 1 5 375 418 1155 4601 483093 0.95 5 10 15 417 460 1197 4643 483135 0.85 6 11 19 421 505 1477 4923 483415 0.75 9 16 26 497 591 2070 6104 510498 0.65 13 23 34 550 665 2247 6979 511373 0.55 9 8 7 6 5 4 3 2 Co-occurrence pattern number minconf minsup 0 1 5 375 418 1155 4601 483093 0.95 5 10 15 417 460 1197 4643 483135 0.85 6 11 19 421 505 1477 4923 483415 0.75 9 16 26 497 591 2070 6104 510498 0.65 13 23 34 550 665 2247 6979 511373 0.55 9 8 7 6 5 4 3 2 Co-occurrence action patterns minconf = 0.95 and minsup = 7 settle allocate calculate process archive enter process permit evaluate level process pick consequent antecedent Evaluation Discussion

• supp and conf values are small

• … the vocabulary contains synonyms

• some action patterns identify process variants

• the number of derived action patterns is

• reasonable

Conclusion

• Contributions

• Action pattern concept

• Action patterns

• derivation

• Validation with

• SAP Reference Model

• Future work

• Automatic action

• derivation

• Similar actions

• clustering

• Advanced algorithms

• for behavioral profiles

• derivation

• Thank you!