Discovering Temporal Work Transference Networks from Workflow Execution Logs

Abstract

Workflow management systems (WfMSs) automate and manage workflows, which are implementations of organizational processes operated in process-centric organizations. In this paper, wepropose an algorithm to discover temporal work transference networks from workflow execution logs. The temporal work transference network is a special type of enterprise social networks that consists of workflow performers, and relationships among them that are formed by work transferences between performers who are responsible in performing precedent and succeeding activities in a workflow process. In terms of analysis, the temporal work transference network is an analytical property that has significant value to be analyzed to discover organizational knowledge for human resource management and related decision-making steps for process-centric organizations. Also, the beginning point of implementinga human-centered workflow intelligence framework dealing with work transference networks is to develop an algorithm for discovering temporal work transference cases on workflow execution logs. To this end, we first formalize a concept of temporal work transference network, and next, we present a discovery algorithm which is for the construction of temporal work transference network from workflow execution logs. Then, as a verification of the proposed algorithm, we apply the algorithm to an XES-formatted log dataset that was released by the process mining research group and finally summarize the discovery result.

1. Introduction

Workflow is a sequence of steps, tasks, events, and those interactions that implement an organizational process for performing repetitive and standardized procedural work in anorganization. Such organizational processes may contain many parts and individuals involved. For example, a simpleonline shopping workflow involves the following steps: First, a customer looks for the product and price, and if it issatisfied, then decide to buy the product. The customer fillsout personal information into an input document for the payment and finally make payment will be completed. Likewise, for enterprise information systems, such as business process management (BPM), customer relationship management (CRM) and enterprise resource planning (ERP) systems, a workflow management system (WfMS), which is capable of automating workflow processes, can be a backbone platform of those systems.

In recent, exploiting and analyzing the historical data of workflow operations have been paid much attention [1-3]. Itenables us to evaluate the performance of workflow processes so that effective adjustments can be made timely to achieve better process performance. In particular, social network modeling and analyzing techniques [4-8] can be applied to workflow processes to uncover social interaction between process workers (performers) and evaluate the performance from the organizational aspect.

In this paper, we propose a technique for discovering temporal work transference network from workflowexecution logs. Temporal work transference networks represent directional relationships between performers whoare responsible for certain preceding and its succeeding activities (a unit of work) in a certain execution case of the workflow process (workcase). These relationships havesignificant meanings of interactions and collaborations between performers in workflow executions and can be applied to a variety of analytics tasks from the organizationalaspect. To this end, we use an execution log dataset formatted in the XES (eXtensible Event Streams) standard [9]. This log format is based on the grammar for tag-based languages aiming at providing designers of informationsystems with a unified and extensible methodology forcapturing systems behaviors by a means of event logs andevent streams that are defined in the XES standard.

During executions of workflow processes, tremendoushistorical data is logged into the log database with detailedinformation, such as activity identifier, performers identifier, organization unit name, relative applications, and timestamp. This information describes the details of the system operations as well as the events that took place out with related entities. The performer is one of the primary entity types that participate in workflow processes and carry outactivities in such processes. A temporal work transferencenetwork consists of nodes of performers and their work transference relationships that can be detected by analyzing execution logs of the corresponding workcase.

With this context, the remainder of this paper is organized as follows. In Section 2, aformal definition of work transference network model that is discovered from the planned workflow process in build-time will be presented. Then, Section 3 provides a formal definition of temporal work transference network model that is discovered from the execution logs of the corresponding workcase. In Section 4, a discovery algorithm of temporal work transference network model will be presented. To validate this work, Section 5 provides experimental results of the discovery algorithm by exploiting an execution log provided by a real company. Finally, this work will be concluded with future works in Section 6.

2. Work Transference Network Model

In this section, we present a formal definition of the work transference network [8] that is formed through the workflow process execution. Basically, procedural activities on a workflow process trigger off work transferences among the performers who are involved in the workflow process. Analyzing work transference network is to acquire human-centered knowledge of workflow executions, and such knowledge can be utilized in essential planning activities in human resource management and decision-making steps.

As a representation of such knowledge, a work transference network model of build-time is defined as Definition 1. A work transference network is represented by a directed graph (or digraph) model to express relationships of work transferences and their associated works among performers who are involved in a corresponding workflow process. Each vertex represents a performer, and each directed edge represents a work transference relationship.

Definition 1 (Work Transference Network Model). A work transference network model of build-time is formally defined as \(\Lambda^{B}=\left(\sigma, \psi, F_{r}^{B}, T_{0}^{B}\right)\), over a set P of performers, and a set A of activities in a workflow model, where :

• \(F_r^B\) is a finite set of coordinators or coordinator-groups connected from an external buildtime model of the work transferene network.

• \(T_0^B\) is a finite set of coordinators or coordinator-groups connected to an external buildtime model of the work transference network.

• \(\sigma = \sigma_i ∪ \sigma_o\) /*Work (Activity) Transferences */

-\(\sigma_o : P \to P(P)\) is a multi-valued function mapping aperformer to its set of (immediate) work (activity) transferrers.

- \(\sigma_i : P \to P(P)\) is a multi-valued function mapping aperformer to its set of (immediate) work (activity) receivers.

• \(\psi=\psi_i∪\psi_o\)/* Work (Activity) Associations */

- \(\psi_i : (P \times P) \to P(A)\) is a multi-valued function returning a set of receiving works (activities) on ordered pairs of performers, \((\sigma_i(o),o),o∈P, \) from \(\sigma_i(o)\) to \(o\)

- \(\psi_o : (P \times P) \to P(A)\) is a multi-valued function returning a set of transferring works (activities) on ordered pairs of performers, \((o,\sigma_o(o)),o∈P\) from \(o\) to \(\sigma_o(o).\) 

(Figure 1) A workflow transference network of buildtime from the library bookacquisition workflow model

Figure 1 shows a graphical representation of work transference network at build-time from the library bookacquisition workflow model. It is composed of eleven performers like \(p_1\)∼\(p_{11}\) and their ordered pairs labeled with the associated works in the activity set, \(A=\{A, C, C, U, U, I, P, G, D, R, T, Z\}\), of the workflow procedure.

3. Temporal Work Transference Network Model

During executions of workflow processes, the logging and auditing components in the workflow execution enginerecord an event on a log database whenever a state of the activity execution changes.

Those logged events are arranged in a form of a temporal sequence of events. This execution sequence of a workcaseforms a workcase event trace and it is also involved with aperforming sequence of the performers (workcase performertrace) who are in charge of the executions of the activities in the corresponding workcase. We can also extract a workcase performer trace from a workcase event trace.

Definition 2 (Workflow Activity Event). Let \(we = (wi, pc,wf, wc,ac,p^*,t,s)\) be an event of activity execution stored as log, where:

• \(wi\) is a workitem (activity instance) identifier.

• \(pc\) is a workflow package identifier.

• \(wf\) is a workflow process identifier.

• \(wc\) is a workcase identifier.

• \(ac\) is an activity identifier.

• \(p^*\) is a performer identifier.

• \(t\) is a timestamp.

• \(s\) is a workitem’s current state, which is one of the statessuch as ready, assigned, reserved, completed, and cancelled.

Definition 3 (Workcase Event Trace). Let \(WT(c)\) bethe workcase event trace of a workcase, \(c\), where \(we_i|we_{i.}wc=c∧we_i. \space t≤we_j.t∧we_i.pc=we_j.pc∧we_j.wf=we_j. \)  

\(\space wf ∧ we_i.wc=we_j.wc∧i<j∧1 ≤ i,j ≤n,\) where formally represents a temporally ordered activity eventsequence of a workcase, which is built through preprocessing the workflow activity events by considering the timestampand the state attributed.

Based on the concepts above, to discover temporal work transferences, we define four types of temporal properties with holding one of the timestamp-origins.

• The scheduled point of time: the event’s timestamp istaken at when the state of a workitem is changed from READY to ASSIGNED. A workflow activity event with ascheduled timestamp, \(we^{(t.s)}⇒ (t=we.t∧s=we.s.∧s=ASSIGNED).\)

• The accessed point of time: the event’s timestamp is taken at when the state of a workitem is changed from ASSIGNED to RESERVED. A workflow activity event with a accessed timestamp,  \(we^{(t.e)}⇒ (t=we.t∧e.=we.s∧e=ASSIGNED).\)

• The started point of time: the event’s timestamp is taken at when the state of a workitem is changed from RESERVED to RUNNING. A workflow activity event with a running timestamp, \(we^{(t.u)}⇒ (t=we.t∧u.=we.s∧u=RUNNING).\)

• The completed point of time: the event’s timestamp is Discovering Temporal Work Transference Networks from Workflow Execution Logs taken at when the state of a workitem is changed fromRUNNING to COMPLETED. A workflow activity event with a completed timestamp, \(we^{(t.o)}⇒ (t=we.t∧o.=we.s∧o=COMPLETED).\)

Based on the temporal properties, we can organize atemporal work transference, TWT(c), by taking the performeridentifier from each of the workflow activity events making up the corresponding workcase event trace. The temporal work transference is transformed into a temporal work transference model, TWTM(c).

Definition 4 (Temporal Work Transference). Let TWT(c) be the workcase performer trace of a workcase, \(c\) :• \(T W T(c)=\left(w e_{\left(p_{1}\right)}^{T(\phi)}, \ldots, w e_{\left(p_{m}\right)}^{T|\phi|}\right)\), where \(\begin{aligned}&w e_{p}^{T|\phi|} | p=w e . p \wedge \tau=w e . t \wedge \phi \in e, u, o \wedge w e_{p}^{T[\phi]}\\&w c=c \wedge w e_{\left(p_{i}\right)}^{\tau}Definition 5 (Temporal Work Transference Network Model, TWTM). A work transference network model ofruntime is formally defined as \(A^R=(\sigma, \psi, F_r^R, T_o^R)\) , over aset P of performer and a set A of activities in a bunch of workcase event traces logged from enacting a specific workflow process, where:• \(F_r^B\) is a coordinators or a coordinator-group linked fromsome external work transference networks of runtime.

• \(T_0^B\) is a coordinators or a coordinator-group linked tosome external work transference networks of runtime.

• \(\sigma = \sigma_i ∪ \sigma_o\) /* Work (workitem) Transferences */- \(\sigma_o : P \to P(P)\) is a multi-valued function mapping aperformer to its set of (immediate) work (workitem) transferrers.

- \(\sigma_i : P \to P(P)\) is a multi-valued function mapping aperformer to its set of (immediate) work (workitem) receivers.

• \(\psi = \psi_i∪\psi_o\) /* Work (workitem) Associations */- \(\psi_i : (P \times P) \to P(A)\)is a multi-valued function returning a bag of receiving works (workitems) on ordered pairs of performers, \((\sigma_i,(o),o),o∈P\), from\(\sigma_i(p) \) to \(p\).

- \(\psi_o : (P \times P) \to P(A)\) is a multi-valued function returning a bag of transferring works (workitems) on ordered pairs of performers, \((p, \sigma_o(p)), p∈P\), from\(p\) to \(\sigma_o(p)\).

4. TWTM Discovery Algorithm

Based upon the definition of TWTM, we develop analgorithm that is able to discover a TWTM from a workcaseevent trace of a single workcase. The algorithm takes XES-formatted execution logs as input data. The XES standard includes an XML schema describing the structure of an XES event log/stream. Moreover, a basic collection of so-called XES extension prototypes that provide semantics tocertain attributes as recorded in the event log/stream isincluded in this standard. Figure 2 represents an XML schema of the XES standard and its core elements are as follows:

• :Root element of an XES log file.

(Figure 2) The XML schema of the XES standard

(Figure 3) Overall procedure of the TWTM discovery

•   : Tag element for a trace object included in aelement. It corresponds to a certain workcase and contains information pertaining to the execution of the workcase from its starting to termination.

• : Tag element for the event object included in aspecific tag element. Each element contains detailed information including activity and performer identifiers, and its timestamp.

Overall procedure of the discovery algorithm is as follows: First, the input XES log file is put into the XES logrepository. The control module reads the input data, passing it to the analysis module. Based on the XES standard structure and data is transferred, the module performs the analysis and disassembly tasks and then transfersintermediate results to the statistic module. Next, the statistic module performs statistical analysis (e.g., frequencies of work trace participation of performers). Finally, the visualization module relies on data generated by previous modules tovisualize the temporal work transference network. To implement this algorithm, we split the job into two sub-steps:

• Step 1: Finding individual performers. At this step, the system performs the parsing data from the XES file, extract each temporal workcase and save it to the system. From each workcase, remove the duplicated performers then add it to a list.

• Step 2: Discovering a temporal work transference network. With the data that has been initialized from step 1, the systems constructs relationships between performers thenfinally visualize the discovered temporal transferencenetwork.

These steps and algorithms can be summarized in Figure 3.

 

5. Experimental Results

In order to validate the TWTM discovery algorithm, weexploit a real log dataset, provided by 4TU [10] and containing execution logs of a loan application process in the anonymous bank company. The summary of the log datasetis presented in Table 1.

(Table 1) Summary of the log dataset.

As an operational example, we applyour discovery algorithm to the dataset and Figure 4 is a visualization of the discovered TWTM of the trace-2 in the dataset.

(Figure 4) Temporal work transference network model discovered from trace-2

For the trace-2, there are 59 events and 10 performersparticipated in the corresponding workcase. As shown in Figure 4, there is a chordless cycle [11] between performers: 10912, 10861, 11203, 11181, 11189, and 11119.

(Table 2) Involvement occurrences of performers for the trace-2

• The entire order of performers attached to its event: 112,112, 112, 112, 10912, 10912, 10912, 10912, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11201, 11119, 11119, 10912, 10912, 10861, 10861,10861, 10861, 10861, 10861, 10861, 10861, 10861, 11203,11203, 11181, 11181, 11189, 11189, 11119, 11119, 11049, 11049, 11049, 11049, 10609, 10609, 10609, 10609, 10609,

The number of turns involved in the performer's event is presented in Table 2 and Figure 5. The workcase (trace-2) was started by the performer 112 and was finished by the performer 10912. For the performer 11201, he/she have participated in the 19 workitems.

(Figure 5) Involvement occurrences of the performers in trace-2

6. Conclusion

The significance of analyzing workflow execution logs forimproving workflow processes has been growing in recent. In this regard, we define the temporal work transference network model (TWTM) that reflects interactions between performers in workflow executions and aims to enable workflow process analyses and improvements from the organizational aspect. Next, we present an algorithm todiscover a TWTM from workflow execution logs.

To verify the proposed algorithm, we implemented the discovery algorithm and applied it to the execution logdataset of the loan application process [8]. Through the demosystem, we confirm that the proposed algorithm is feasible and can discover TWTM from given execution logs.

However, the correctnessof discovered TWTMs remains as a challenge. Accordingly, to discover exact or moreprecise TWTMs, we are planning to study the following in the future:

• Handling complicated control-flow patterns (e.g., loop).

• Developing a comprehensive analytics system forvisualizing TWTMs and providing those statistics results.

• Proving the effectiveness and utility of the TWTM discovery and analytics with applying to real-worldworkflow processes.