Social Process Mining (SPM) combines the research fields Process Mining (PM) and Social Collaboration Analytics. The SPM-Cockpit is a prototype that can detect and analyze Collaboration Patterns in event logs of Enterprise Collaboration Systems (ECS) semi-automatically. We use this tool to investigate activity patterns that occur while users work collaboratively in ECS. To do so, we adapt methods from Process Mining, Frequent Subgraph Mining, and Graph Clustering and bundle them in the SPM-Cockpit. The prototype is a first step towards investigating, understanding and categorizing collaboration activity in ECS automatically.
In recent years, many workers have shifted to a remote workplace [
Process Mining (PM) is an established research domain that analyzes event logs extracted
from software systems [
The SPM-Cockpit aims to detect patterns of collaborative activities of users in ECS. SPM is the acronym for Social Process Mining, which combines the research fields of Process Mining and Social Collaboration Analytics in ECS. As there are many steps and several methods involved in the process of SPM, we need a tool that bundles these methods and guides the users in their application to ease access. Thus, the SPM-Cockpit provides tailored methods for preprocessing log data, for the discovery of process models, frequent subgraph mining and graph set clustering algorithms for the detection of collaboration patterns, and for creating collaboration pattern repositories (explained in the following section). Furthermore, the cockpit guides the users through the SPM process. The outcomes of the SPM-Cockpit are collections of collaboration patterns (CPs) that help to describe collaboration. CPs are subsections of process models that can be found frequently in process models mined from ECS logs. The primary target user for the SPM-Cockpit is an ECS analyst or a researcher who aims to improve the collaborative work processes of ECS. Thus, s/he can use the cockpit to mine CPs from ECS process models, which describe the collaborative interactions of different users with social documents. These outcomes illustrate the typical collaboration activities in ECS. Eventually, the CPs can be used as building blocks that can be embedded in Business Process Management environments to improve collaborative business processes and ECS.
In the remainder, we describe the SPM-Cockpit and its features in Section 2. Section 3 describes the maturity of the tool. Finally, Section 4 concludes the demo with an outlook to future work with further feasible improvements.
The SPM-Cockpit is a web application that guides the user through the process of SPM. It is built
with the Django web-framework2 and uses the following complementary technologies: a
relational database (MariaDB®) for storing configurations and user sessions, a Redis® database3 for
caching traces, a Neo4j®4 graph database for storing the collaboration pattern repositories, and
Celery5 for distributing computationally expensive tasks with task queues. In addition, the
application uses the process mining library pm4py [
Cache
Message Broker
Celery Messages Cache
DB access DB access application
DB access DB access
The BPMN model (Figure 2) shows the underlying process of the application with the SPM functionalities as activities and steps. If the results of an intermediate step do not satisfy the expected outcome, the user can step back and reconfigure the parameters and redo the corresponding activity (e. g., redo the process mining discovery with another algorithm). The SPM-Cockpit divides the proposed analysis into six main activities: 1. Importing an Event Log, 2. Preprocessing & Process Mining Configuration, 3. Exploring Process Models, 4. Frequent Subgraph Mining, 2 https://www.djangoproject.com (last access: 24th of July, 2023) 3 https://redis.io (last access: 24th of July, 2023) 4 https://neo4j.com (last access: 24th of July, 2023) 5 https://docs.celeryq.dev/en/stable (last access: 24th of July, 2023) it -ckopC PSM
ECSEXvEeSntLog
Importingan EventLog
PPrCreoopcnreofiscgseusrMasitinnioignng& ProEcexspsloMrinogdels
RedisCache
ProceosksayM?odels no
GraphDatabase Frequent Subgraphsokay? Subgraph Mining no GraphSet Clustering
Clustersokay? no
Managing Colaboration Patern Repository 5. Graph Set Clustering, and 6. Managing Collaboration Pattern Repositories. The entry point of the SPM-Cockpit is the landing page with an overview of these SPM steps (Figure 3a). The following paragraphs explain each activity and describe their purpose, functions, and in- and outputs.
The first step is Importing an Event Log. ECS event logs have some special properties that are
relevant for the further SPM process. We assume that the baseline case (the process instance) for
such an event log is based on the so-called workspace (i. e., a particular working environment in
an ECS that was created, for instance, for a project). For example, all events that were fired in one
workspace belong to one trace in the event log. We further require the events to have the
following attributes: the associated social document, the executing user (resource), and the timestamp
when the event was fired. As eXtensible Event Stream (XES) [
After the upload is finished, the tool starts the second activity (preprocessing) with the default
parameters. Thus, the second step, Preprocessing & Process Mining Configuration, triggers the
preprocessing phase and defines how the process models should be mined. To detect which persons
are working together on which social documents, the traces of the workspaces are further split
into sub-traces. The intent is that we split each trace (the workspaces) into smaller traces based
on those social documents that are jointly worked on by several users. Then, for each sub-trace,
we identify ECS users who have worked in the same timeframe (of the trace) on the same
documents and group those traces with direct or indirect overlaps in common users and timeframes.
Thus, we create multiple sub-logs (the groups of sub-traces) based on the collaborative activities
of the ECS users. Then, for each of these sub-logs, we can mine process models that express the
collaboration. The process models are generated based on the PM configuration. For now, the
cockpit user can decide whether to discover Directly Follows Graphs [
The intermediate step, Exploring Process Models, lets the cockpit user investigate whether the generated process models are sufficient for further analysis. Therefore, the SPM-Cockpit provides visualizations of the generated process models with basic navigation, pan and zoom functionality. The user can inspect the process models and step back if s/he has concerns regarding the PM configuration.
The Frequent Subgraph Mining step uses the previously generated process models and
searches for patterns using Frequent Subgraph Mining (FSM) [
During the Graph Set Clustering step, similar patterns (subgraphs) are clustered into groups. Such groups then represent similar collaboration patterns rather than identical ones. The reason is that we assume that a typical collaboration scenario does not always look the same but rather similar. These groups/clusters can be viewed in the next step in the Collaboration Pattern Repository. The user can choose a hierarchical clustering algorithm or a partitioning clustering algorithm. Where the first requires the analyst to define the number of clusters, the latter finds the number of clusters based on the dissimilarities of the given graph set6. Again, based on the results, the user can redo the clustering process with a new configuration.
In the last step, Managing Collaboration Pattern Repository, the user can investigate the resulting clusters, which we store in a Collaboration Pattern Repository (CPR). Each cluster consists of a set of patterns (subgraphs) and can be visualized with a representative graph (e. g., the initial graphs from the clustering step). Additionally, the user can add a name and a description for the respective cluster as they represent similar collaboration patterns. The analyst can now use these patterns for further analysis that can provide insights into how the collaboration was conducted in the ECS. An example of two found patterns is illustrated in Figure 3b. These show the collaborative activities of users working on a wiki page and a file. As the CPRs are stored in a graph database (Neo4j), other existing CPRs can be loaded and inspected as well. Furthermore, the SPMCockpit provides the functionality for down-/uploading the CPRs.
We deployed one instance of the SPM-Cockpit7 on a server. Furthermore, we published the related user documentation of the SPM-Cockpit8 and the technical documentation of the underlining Python library9. We provide some ECS event logs, which one can select in the first step as sample data (instead of uploading an event log) directly in the SPM-Cockpit. We successfully applied the SPM-Cockpit with these event logs, which we demonstrate in the SPM-Cockpit documentation as a walkthrough video tutorial10.
The SPM-Cockpit is a tool for the analysis of ECS event data and provides a bundled toolset for the discovery and investigation of collaboration patterns. Thus, this tool extends the analysis of 6 Additional clustering parameters can be defined by the user, but we refer here to the documentation. 7 https://w3id.org/spm/cockpit (last access: 24th of July, 2023) 8 https://w3id.org/spm/docs/cockpit (last access: 24th of July, 2023) 9 https://w3id.org/spm/docs/colpadef (last access: 24th of July, 2023) 10 https://w3id.org/spm/docs/cockpit/tutorial & https://youtu.be/4PYqSRKEPDw (last access: 24th of July, 2023) unstructured processes in (social) document-orientated systems. For this purpose, we apply algorithms from the PM domain, graph-based algorithms, and clustering algorithms. For future improvements, we aim to: • improve the preprocessing by including a parameter for temporal overlaps of the sub-traces to refine the sub-log generation. • include further process discovery algorithms. • extend subgraph discovery by implementing a relaxed Frequent Subgraph Mining algorithm that also finds subgraphs that are similar to the search pattern rather than being isomorphic. This way, similar subgraphs that would normally have a too low support in a “classic” FSM will not be missed. • implement and evaluate further clustering algorithms to find the best one for SPM. • apply existing CPRs to analyze unseen ECS event logs, while calculating metrics based on the patterns and the unexplored event log.
While we focus on ECS event data in SPM, further systems logs may be investigated with the help of the SPM-Cockpit in future research.
This work was partly funded by the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) – project number 445182359.