There has been a huge body of research in order to reduce manual efforts in creating executable process models through the automated discovery of process models from the event logs created by information systems. Regarding activity-centric processes, such event logs comprise case ids and events related to the execution of process activities. However, there exist alternative process management paradigms, such as object-aware processes, for which existing algorithms fail to discover a sound model. These algorithms do not treat data as first-class citizens, but solely rely on the information from event logs. In consequence, existing discovery algorithms are insufficient for discovering object-aware processes. To address this issue, discovery algorithms need to consider additional data sources (e.g., existing forms). This paper discusses the need for dedicated discovery techniques in object-aware processes.
Despite the many mining approaches that exist for activity-centric processes,
adequate support for discovering data-centric process models, e.g., in the context
of artifact-centric processes [
To understand the nature of the problem at hand, a short introduction into
data-centric and object-aware process management becomes necessary.
PHILharmonicFlows, our approach to data-centric processes, introduces the concepts of
objects, object behavior, and object interaction. For each business object present
in a real-world business process, one such object exists. The latter comprises
data, represented by attributes, and a state-based process model describing the
object behavior in terms of an object lifecycle model. When data becomes
available during runtime, this enables transitions between the various states of the
lifecycle process, i.e., execution is data-driven. In the e-learning system
PHoodle, a practical application of the PHILharmonicFlows system [
Lecture 1:n Download 1 : n 1 : n
1 : n 1 : n
Exercise #1
Exercise #2
AELtifxreiecbyructcleeAissseiEgExnedmrcieitnset::SFSttiurlidenegsnt SuFilebPs:omFiilnetissSAssuiigbonmmPnoeiiFnFntFetet#:ste:TeddIeub1bndtaatdoeccrkgbke==arcftkarlusee Fe dback:PBFoaaoslilseedd
AELtifxreiecbyructcleeAissseiEgExnedmrcieitnset::SFSttiurlidenegsnt SuFilebPs:omFiilnetissSAssuiigbonmmPnoeiiFnFntFetet#:ste:TeddIeub2bndtaatdoeccrkgbke==arcftkarlusee Fe dback:PBFoaaoslilseedd
AELtifxreiecbyructcleeAissseiEgExnedmrcieitnset::SFSttiurlidenegsnt SuFilebPs:omFiilnetissSAssuiigbonmmPnoeiiFnFntFetet#:ste:TeddIeub3bndtaatdoeccrkgbke==arcftkarlusee Fe dback:PBFoaaoslilseedd
Process discovery summarizes techniques that leverage information from event
logs to discover process models [
Towards the Discovery of Object-Aware Processes
Research direction
In our PHILharmonicFlows framework, an object-aware process consists of a data
model, one lifecycle model for each object, and a coordination process enforcing
constraints regarding object interactions [
The first step during process discovery is to identify objects, including their attributes and relations. Note that the structure of a normalized relational database, to a certain degree, is comparable to a data model, which offers the opportunity to discover the data model from the structure (i.e., the create table statements) of a database. Each table in the database may, but does not have to correspond to an object in the data model, whereas columns of a table represent the attributes of an object. One-to-many relations between tables can be used to identify relations between the objects of a data model. Additionally, relations can be used as an indicator if a table corresponds to a correct object.
After discovering the data model, the object lifecycles need to be discovered in the second step. Based on the attributes from the data model, lifecycle discovery shall deliver object states as well as the transitions between them. In general, a lifecycle process may enter another state, if all necessary data (i.e., attribute values) are available. In particular, lifecycle states cannot be discovered from event logs, whose entries solely refer to activities due to the mismatch between states (i.e., defined by attributes) and cases (i.e., a collection of activities). To tackle this mismatch, discovery algorithms for object lifecycles, suitable event log preprocessing (e.g., splitting an event log into event logs for each object), and additional data sources (e.g., forms of existing information systems) need to be considered as well during the discovery process.
The third step in discovering an object-aware process is to unravel the coordination of interactions between objects (e.g., a submission may only be created if the corresponding exercise is in state published). As object interaction can only be discovered with the data model and the lifecycles present, their discovery is a secondary problem for now. 4
Conclusion This paper discusses the need for spending research efforts on the discovery of object-aware process models. As major advantage, the discovery of an objectaware processes allows to identify the underlying logic of a process. Finally, due to the strong linkage between process and data in object-aware processes, it is possible that not every aspect of each element (i.e., data model, lifecycles, and coordination) may be discovered from the presented data sources and, therefore further research is of utmost importance.