Compared with the established and well known BPMN approach, the new approach introduced with this paper appears to be odd, but despite this coincidence, ODD-BP stands for ontology-and data-driven business process approach. The motivation arises from the research about knowledge intensive processes (KiPs) with its data-oriented character [5] and the general search for AI-support within business processes. Established approaches often place data as a third class citizen into the process definition, sometimes expressed as an afterthought [4] within a classical control-flow oriented model. KiPs are knowledge-and datacentric and require flexibility at design and run-time [3], especially regarding their major resource, the knowledge workers. They should be supported by offering opportunities rather than restrictions, an accompanying system should deliver choices and recommendations and access to relevant information to accomplish a contribution during the process execution [3,5].
Artificial intelligence (AI) in all its facets and across the different kind of technologies are discussed in a wide range of use-cases as well as it is in the focus of research in the BPM-context. No matter if an AI contribution is delivered by the workflow system itself or by an external agent, when it comes to a division of labour between human-and cyber actors, the process must be described without ambiguity and understandable for all process participants. An ontology is the perfect tool to achieve this requirement and can be used for a semantic process definition according to Fellmann [24].
The combination of both leads to an ontology and data driven ODD-BP approach as it is described in this paper. The data-driven character is formed by a metamodel defined in the base-ontology as the core of the conceptualization. Aligned with our research project SEMANAS 3 , we examine the demands of knowledge intensive process in the domain of agricultural grant applications and expand the base-ontology by a domain-ontology focused on this specific use-case.
Section 2 is referring the foundations of our work, while section 3 introduces the metamodel for our new approach. The application of the metamodel is described in section 4, while additional transition rules expand the conceptualization to ensure a valid execution of process instances. Section 5 gives a short insight into our current development of a process design-tool. With the conclusion in section 6, we give an outlook about the possibilities and advantages of the ODD-BP approach and our future work.
Business Process Model and Notation (BPMN) is currently the de facto standard for designing and describing business processes world-wide. In the center of this approach reigns a control-flow coordination of process steps (activities). A less restrictive, but still activity-centric perspective is provided by constraintbased approaches [19], which allow flexibility in a scalable manner. Alternatively, there are several approaches with the intention to gain flexibility based on the control-flow principles [23,16]. Despite of the consideration of data-flow in such processes, the data is just integrated in a kind of an afterthought [4,11]. Opposed to this, knowledge-intensive processes (KiPs) are usually barely structured and their execution is driven by user decisions and business data. Previous research has shown [25,2,17] that an activity-centric perspective is not sufficient to support knowledge-intensive processes.
With view to these insights, several new approaches were brought up during the last decade, putting the data into the center. The case handling paradigm [25] elevated the result of a process (case), reflected by its data objects; activities do not longer drive the process but serve the outcome. For more complex scenarios with the need of abstraction capabilities, object-awareness approaches refined the case handling concept [26]. With business artifacts [1,2], CorePro [17], and PHILharmonicFlows [15] there are even more approaches to mention, which underlines the importance of data-centric approaches for knowledge intensive processes.
Beside the examinations of different workflow models and principles, the possibilities of semantic information systems in the domain of BPM are matter of current research as well. S-BPM introduced a subject-oriented modeling scheme, where sentences with subject, predicate, object are used to describe general interactions between process actors [8]. This profound methodology addresses the communication aspects between process actors, but offers no specific strategies regarding the demands of KiPs. The initiative of WSMO (Web Service Modeling Ontology) uses ontologies to formalize the interoperability of web services [7] and thus targets a service orchestration purpose. Some work considers how a formal semantic can be utilized for process validation [20] or optimization [9] purposes. Opposed to this, Thomas and Fellmann [24] introduced an ontology based representation of business processes in which process elements are assigned to ontology classes to define a control-flow oriented metamodel. Further research also uses semantic process modeling to define control-flow oriented approaches like [12,13]. In general, the research about semantic process definitions is motivated by the reduction of ambiguity [6,24] and the opportunity of inferencing new knowledge [14] on base of the existing process knowledge.
The research about semantic formalizations within the field of business processes defines the foundation for our work, but to the best of our knowledge, no research addresses a semantic process modeling principle for a data-oriented workflow approach to support the definition and execution of knowledge intensive processes. The general idea for this approach was already introduced through our precedent work [22,21] and will be carried forward with this paper.
According to a wide range of publications [10,18], a metamodel defines "the frames, rules, constraints, models and theories applicable and useful for modeling a predefined class of problems." Knowledge intensive processes (KiPs) and their specific demands [5] can be considered as such a predefined class of problems. We define the ODD-BP model, a workflow metamodel aligned to the requirements of KiPs, utilizing the underlying ontology to provide semantic and data-oriented process definitions.
Any workflow management system follows a metamodel, usually implementing one of the workflow-approaches mentioned in section 2. From a most fundamental perspective, all WF-approaches have in common, that they use a set of activities to achieve a specific goal. The metamodel defines the kind of process elements and their possible interactions and based on this model, a process definition (PD) is specified for each kind of process goal, acting as a blueprint for process instances (PI) which can be executed to achieve a specific goal of a certain kind.
All established workflow approaches follow this general WfMS-structure built on a metamodel, process definitions and process instances (Fig. 1). The knowledge store (KS) is the combination of the ontology (T-Box) as the conceptual fundament and the triple store (A-Box) as the data storage. The base ontology defines the metamodel 1 of the ODD-BP approach, it is domain independent and will be introduced in detail in 3.2. The domain ontology expands the conceptualization by defining general valid concepts and relations of a domain. Further ontologies can expand the conceptual knowledge, like a document ontology, and form together the enterprise ontology, the fundament 2 of all process definitions and process instances. A process definition or a process instance is defined by a set of linked individuals according the concepts and relations of the metamodel. Such a set represents a process graph and is stored as triples 3 in the A-Box.
The base-ontology defines all concepts and relations to build the metamodel of the ODD-BP approach. The most fundamental artifacts of a process are Tasks, Dataobjects, Documents and Actors. Individuals of these concepts can be connected with an individual of the class Process through a contains or involves relation, as it is shown in Figure 2.
The mentioned artifacts are usually represented in one or another way in all workflow approaches. The specific character of a metamodel is manifested by the kind of relations between these artifacts and in this case the data-oriented character of the ODD-BP approach is created by the relations between Tasks at the one side and Documents, Dataobjects and Attributes on the other side.
According to this metamodel, a Document can be demanded by a Task as input or a Task can produce a Document as the outcome of its execution. Analog to this, a Dataobject or an Attribute can be required by a Task as input or a Task can deliver such an element as output. The deeper meaning of Dataobject and Attribute will be explained more in detail in the following, but the general importance of data for a process execution is obvious.
Unlike these direct relations between the named concepts, the possible relations between Actors and Tasks are not defined in a direct manner, but through specialized concepts. This allows a precise definition of different kind of Actors and Tasks and their individual relations. Corresponding to this, a cyber actor (labeled as Agent) can perform a System Task. A Role can just be assigned to a User-Task, while a User can execute such a User-Task, if the User is allowed to play the corresponding Role.
All in all the metamodel defines, how a process can be designed and executed and these general rules are the same for process definitions PD and process instances PI, which are defined as a specialization of the concept Process. A process is modeled and described by individuals of the introduced concepts and by links between these individuals according the relations of the base ontology.
The general data-orientation of the metamodel can be seen by the manyfold relations between Task and the data-carrying elements Dataobjects, Attributes and Documents, in certain ways similar to artifact centric approaches [1,2,4]. However, the ODD-BP approach is not only placing data into a more central position, the data is integrated with the intention to drive a process. Usually an information system organizes data about the real world with entities and relations. With view to databases, entities are managed as entries into a table, while a knowledge store is managing entities as individuals of a certain class. Such an individual represents an object of the real world, while its objectcharacteristics are stored as data-properties of the individual. This realization lacks in expressiveness, as the knowledge store can not express dependencies between tasks and data-properties and thus, the data-properties could not be used to drive the process.
This leads us to a conceptualization in which object-characteristics are represented through a separate concept, Attributes. They can be understood as key-value pairs, while an individual of this type is representing a single characteristic of an entity. With view to the example shown in Fig. 3 Assuming that a system task Is Adult can decide autonomously if a Person can be seen as adult or not. The relation between dataobject and task would usually just be expressed at an abstract and more general level 1 . Through the additional conceptualization, the relation between the system task Is Adult and the Attributes 2 can be modeled more in detail. This offers a more specific process definition which allows to deduce the executability of tasks by an inference engine using the linked (input-)attributes with a required by relation. Beside the pure executability, such activities can deliver a more or less important process contribution to achieve a process goal which can be deduced using the linked (output-)attributes with a delivers relation.
Beside the entities itself, an information system also allows to express relations between entities. The base-ontology (Fig. 2) defines two different basetypes to express such relations. The has a relation is intended to define any kind of global valid relations between entities. Within a further conceptualization through a domain ontology, specializations of the has a relation can be used to form an enterprise information model, equivalent to an ER-model of a database.
The composed by relation is the prototype of a more process dependent combination of entities, not valid from a general perspective, but valid and required within a process context. It can be used to combine different kind of dataobjects for a process definition and process execution.
Data can come from different sources like an information system, from a user or cyber actor (through a process or system task) and by documents, which usually means data-exchange with external systems or users. From this perspective, a document can provide dataobjects (Fig. 3) within a process context. According the introduced design to represent entities, a document can also provide attributes of dataobjects. These provide relations are also introduced by the base ontology (Fig. 2) and build the foundation for a semantic integration of documents into a process context. This topic is object of our current research and expands the possibilities of the ODD-BP approach even further.
The introduced metamodel defines the vocabulary, which allows a process designer to express a process definition for a certain process goal. The data of process definitions and process instances as well as any data of the information system is part of the knowledge store and is persisted as triples in the triplestore, the A-Box. Since any process-element within a process definition is just used to form a process template, these elements are placeholders and with view to a process instance, such placeholder elements will be replaced by meaningful elements during the process execution. To express this, the base-ontology (Fig. 2) also defines a placeholder concept. Any process-element of a PD is represented by an individual assigned to one of the introduced concepts (task, dataobject, attribute, document) and additionally assigned to the placeholder class. This serves the inference mechanism to deduce the executability of activities as we mentioned before and which will be addressed in detail by a separate paper.
A process instance is initially nothing else than a copy of a process definition with all its placeholder elements. Along the execution, the process instance performs a gradual transition from a process description defined only by placeholder elements to a final process state, where some or all process elements are meaningful elements. This gradual transition must follow some rules, but since the process descriptions of PD and PI are defined by a set of individuals in the A-Box, the rules can not be expressed using the conceptual layer of the knowledge store. In the following we will introduce these rules which define valid structural changes between a modeled process definition and corresponding executed process instances.
A PD defines a template to achieve a process goal like an application for a group event. In such a PD, a single individual of the classes dataobject and placeholder is representing the applicants, not knowing how many applicants will finally participate in a PI. To express this aspect of a multi-instanciation, the metamodel must offer the option to define the cardinality for each process element (PE). Since the placeholder PEs will be replaced during the process execution, the right place to persist the possibilities of the cardinality is at the link between the individuals of the process p and the process element pe. For this purpose, the contains relation is expanded by an annotation, which allows each link of this type to define the cardinality by a list with 2 numbers (n, m).
-n ∈ N 0 defines the minimum occurrence of a process element, while n = 0 defines an optional occurrence within a PI. -m ∈ N 0 defines the maximum occurrence of a process element, while m = 0 defines an unlimited occurrence within a PI. -In the following, the cardinality will be expressed with an additional label at the contains relation as:
pe
The multi-instanciation only defines the possible occurrence of PEs. It tells nothing about the way, how such PEs interact with each other. This requires a further extension to express the intended transitions between PEs.
According the metamodel, Tasks, Documents and Dataobjects are connected with a Process through the contains relation. The interplay between individuals of these three concepts are defined by the five relation types provides, demanded by, produces, required by and delivers. Analog to the ontology, two connected process elements can be seen as a process statement with subject, predicate and object, while subject and object are individuals of the three mentioned concepts and predicate is a link of the five mentioned relations. Such a process statement defines an action (according the predicate) from a subject to the object. Since subject and object can be defined with a different cardinality, the predicate must be extended with an annotation how the multi-instanciation can be processed to perform the transition from placeholder PEs to meaningful PEs. We can differentiate in three transition rules:
-expand: Each subject can result (along the predicate) in one or many objects.
-maintain: Each subject can result (along the predicate) in exactly one object.
-join: All subjects can result (along the predicate) in exactly one object. -In the following, the transition rules will be expressed with an additional label at the predicate as: pe As an example, Figure 4 shows a small process segment of a group application with different cases of multi-instanciation and transition rules. The left side presents the process definition, where the links of the contains relation are annotated with the definition for a multi-instanciation, while the other links are annotated with the transition rules. The PD defines, that exactly one (1:1) Group Application is allowed, it allows at least 2 and a maximum of 6 Persons (2:6), any number of Is Adult tasks and just exactly one task Group Total Age. These four process elements are linked through provides and required by relations with the three different transition rules (expand, maintain, join).
The right side presents a corresponding process instance, where all process elements are transformed from placeholder PEs to meaningful PEs, according the introduced rules. The Group Application has provided two Persons, which fulfills the demand of the contains relations and follows the expand transition rule. It is required, that each Person must be an adult and the maintain transition rule is defining, that for each Person an own Is Adult task must be executed. Additionally, the total age of the group is needed, which is why both Persons are linked with just one Group Total Age activity according the join transition rule.
The metamodel defines the vocabulary to describe a process and through the introduced rules regarding multi-instantiation and transitions, also the semantic dependencies between process elements can be expressed. For a practical use and to design and execute ODD-BP definitions and instances, a modeling tool is required.
As a first step towards a POIS, we started the development of a graphical toolset on a web-based client-server architecture. It aims to support a process designer by utilizing the metamodel to turn the ontological restriction into an easy to use graphical user-interface. Beside this, it also sets the outer ontological restriction into action. The editor is highly generic since the possibilities to form the process is ascertained through the ontologies. Beside the visual interpretation of the process definition, the tool is taking care for a consistent process model by realizing the outer ontological restrictions. As an example: As soon as the user is deleting the linkage between Is Adult and Person, the editor will also remove all links to attributes of the dataobject Person. The ontology defines no limitations for this case, which is why the editor has to fill this gap to prevent inconsistency.
AI technologies in general can deliver a wide range of contributions within the field of business processes and even semantic information systems can be used in different ways. The introduced ODD-BP approach combines the principles of semantic process definitions [24] with a metamodel which implements a declarative and data-oriented process character. Thus, it reduces ambiguity and supports the division of labour between human and cyber actors and takes advantage of a data-oriented approach according the demands [5,3] to design and execute knowledge intensive processes. Through the renunciation of control-flow principles and our focus to a descriptive process model, we gather the advantages of none imperative approaches [11,19], regarding flexibility during the process execution.
One central motivation is to utilize the ontology and the data-oriented metamodel to drive the introduced approach, which leads us to the acronym ODD-BP approach. This requires a precise definition of the interplay between data and activities and we have shown, that the conceptualization of Attributes through the metamodel expands the expressiveness of a process definition and defines the base for different kinds of cyber process contributions. Within a separate paper we will show, that a process definition on base of the ODD-BP approach can be used to deduce the executability of activities. Further more we will show, that even the relevance of an activity can be deduced according its contribution to reach predefined process goals and process milestones.
There is a wide range of further possibilities to take advantage of the ODD-BP approach. As such, the explainability of the inference while deducing executable tasks could be utilized to adapt a process instance according an identified problem. The semantic process definition could also be used for an adaptive process visualization as it was already introduced [22,21] and which is object of our ongoing research. Finally, the practical use of our new approach within different knowledge intensive application scenarios must be examined as well.