DiscopFlow: Organizational Workflow Mining

• DiscopFlow is a specialized workflow mining tool that extracts and analyzes organizational structures and communication protocols from enriched business process logs.
• It leverages modular analyzers and FIPA-ACL performatives to detect hierarchies, negotiation patterns, and protocols such as contract-net, auction, and voting.
• The tool integrates multiple modules with a central database to support extensible analysis and plans for XML log conversion for enhanced interoperability.

DiscopFlow is a workflow mining (WM) tool designed specifically to uncover and analyze organizational structures and interaction protocols embedded within business processes. Distinct from prior WM systems, DiscopFlow leverages enriched workflow logs incorporating FIPA-ACL performatives, enabling extraction of social and communicative dynamics such as hierarchies, coalitions, federations, markets, and negotiation patterns including contract-net, auction, and voting protocols. Its architecture employs modular analyzers to separately address organizational structure, protocol detection, informational perspective, graphical representation, and basic performance metrics, grounded in a formal meta-model extending classical workflow logs by actor-to-actor interaction events (Abdelkafi et al., 2012).

1. System Architecture and Modules

DiscopFlow comprises seven integrated software modules and a central workflow-log database (Oracle 10g), organized as follows:

Module	Core Function	Status
LogGenerator	Create/populate enriched log per workflow meta-model	Implemented
LogConverter	Export relational logs to XML	Under development
OrgStruct_Analyser	Detect and classify organizational structures	Implemented
InterPro_Analyser	Detect organization-level protocols	Implemented
Info_Analyser	Discover document flows per activity	Implemented
AGR_Analyser	Visualize agent-group-role (AGR) relationships	Implemented
Performance_Analyser	Analyze activity and actor performance	Under development

All analytic modules connect to the relational database via JDBC. Log generation precedes analysis; subsequent modules read the enriched event log and produce structural, informational, or graphical outputs. The future LogConverter will provide tool-independent XML export for interoperability and further processing (Abdelkafi et al., 2012).

2. Workflow Log Meta-Model

The DiscopFlow meta-model extends conventional process-event logs by capturing explicit communicative actions between actors. The model is characterized by:

• $P = \{p_1, \dotsc, p_n\}$: Business processes
• $I_p = \{i_1, \dotsc, i_k\}$: Instances for process $p$
• $E = \{e_1, \dotsc, e_m\}$: Log events
• $A, R, U, D, M$: Sets of actors, roles, organizational units, documents, FIPA-ACL performatives

Each event $e \in E$ is a 7-tuple:

$$e = (\operatorname{inst}(e), \operatorname{ts}(e), \operatorname{act}(e), \operatorname{doc_{in}}(e), \operatorname{doc_{out}}(e), \operatorname{sender}(e), \operatorname{receiver}(e), \operatorname{perf}(e))$$

where:

• $\operatorname{inst}(e)$ identifies the process instance,
• $\operatorname{ts}(e)$ is a timestamp,
• $\operatorname{act}(e)$ is the activity,
• $I_p = \{i_1, \dotsc, i_k\}$0 are input/output documents,
• $I_p = \{i_1, \dotsc, i_k\}$1 are actors or ‘System’,
• $I_p = \{i_1, \dotsc, i_k\}$2 is the FIPA-ACL performative.

Actor-role and actor-unit mappings are represented as functions:

• $I_p = \{i_1, \dotsc, i_k\}$3
• $I_p = \{i_1, \dotsc, i_k\}$4

The meta-model captures organizational and informational context, enabling analysis of “who does what” (via activity, role, unit), “with which documents,” and critically, “by which communicative acts” (Abdelkafi et al., 2012).

3. Organizational Structure and Interaction Protocol Discovery

3.1 Structure Detection

The OrgStruct_Analyser builds a directed multigraph $I_p = \{i_1, \dotsc, i_k\}$5 on actors, labeling edges $I_p = \{i_1, \dotsc, i_k\}$6 with multisets of exchanged performatives. It implements pattern-based identification of:

• Strict Hierarchy: $I_p = \{i_1, \dotsc, i_k\}$7 with $I_p = \{i_1, \dotsc, i_k\}$8 'delegate', $I_p = \{i_1, \dotsc, i_k\}$9, $p$0, with no reverse delegation, then $p$1 is a strict superior–subordinate relation.
• Relaxed Hierarchy: If delegation occurs bidirectionally, $p$2 have a relaxed hierarchy.
• Federation: Symmetric exchange of 'cfp' and ‘propose’ on the same activity, no persistent dominance.
• Coalition: Groups mutually propose for an activity, without clear leader.
• Market: Calls for proposal issued to a pool with winner selection denotes market dynamics.

The pseudocode iterates over $p$3, counting delegation events to detect strict versus relaxed hierarchies with time complexity $p$4, where $p$5 and $p$6 (Abdelkafi et al., 2012).

3.2 Protocol Identification

The InterPro_Analyser detects negotiation protocols via pattern-matching on performative sequences per activity instance $p$7:

• Contract Net: Subsequence matches $p$8 $p$9 $E = \{e_1, \dotsc, e_m\}$0-$E = \{e_1, \dotsc, e_m\}$1-$E = \{e_1, \dotsc, e_m\}$2 $E = \{e_1, \dotsc, e_m\}$3.
• Auction: Pattern $E = \{e_1, \dotsc, e_m\}$4 $E = \{e_1, \dotsc, e_m\}$5 $E = \{e_1, \dotsc, e_m\}$6-$E = \{e_1, \dotsc, e_m\}$7.
• Vote: $E = \{e_1, \dotsc, e_m\}$8, multiple $E = \{e_1, \dotsc, e_m\}$9, then $A, R, U, D, M$0-$A, R, U, D, M$1.

Pattern detection operates in linear time with respect to the length of $A, R, U, D, M$2 (Abdelkafi et al., 2012).

4. Use of FIPA-ACL Performatives

DiscopFlow operationalizes each FIPA-ACL performative as an indicator of social intent:

• delegate: Authorization/subordination
• cfp: Initiation of negotiation
• propose/bid: Offer submissions
• accept-proposal / reject-proposal: Outcome resolutions
• inform: Notification

Sequences and distributions of these communicative acts, when aggregated per actor pair or case, allow reconstruction of both macroscopic organizational forms and microscopic negotiation protocols in process enactment (Abdelkafi et al., 2012).

5. Empirical Example

A segment from the water distribution crisis process log illustrates DiscopFlow’s event representation:

1. (Mahdi → System: Execute(Investigation report establishment))
2. (Mahdi → Salim: Delegate(Alarm triggering))
3. (Salim → Mahdi: Inform(Alarm triggering))
4. (Malik → Sami: Cfp(Analysis samples))
5. (Malik → Amal: Cfp(Analysis samples))
6. (Amal → Malik: Propose(Analysis samples))
7. (Sami → Malik: Propose(Analysis samples))
8. (Malik → Amal: Accept-proposal(Analysis samples))
9. (Malik → Sami: Reject-proposal(Analysis samples))
10. (Amal → System: Execute(Analysis samples))

From lines 2–3, DiscopFlow infers a strict hierarchy between Mahdi and Salim. Lines 4–9 instantiate a contract-net protocol among Malik, Amal, and Sami, with Malik as initiator and Amal/Sami as responders, matching the contract-net pattern in negotiation (Abdelkafi et al., 2012).

6. Comparison with Existing Workflow Mining Tools

DiscopFlow is evaluated against ProM, InWolvE, and WorkflowMiner according to criteria:

Feature	ProM	InWolvE	WorkflowMiner	DiscopFlow
Log Conversion	Yes	Yes	No	Yes
Informational Perspective	Partial	No	No	Yes
Organizational Discovery	Partial	No	No	Yes
Delta/Performance Analysis	Yes	No	No	Planned

DiscopFlow is the only system to provide comprehensive organizational structure and interaction protocol discovery by leveraging FIPA-ACL performatives. No precision/recall metrics are reported, and performance and process-integration modules remain under development (Abdelkafi et al., 2012).

7. Contributions, Open Issues, and Future Directions

DiscopFlow advances workflow mining through:

1. Demonstration that Multi-Agent System (MAS) abstractions can be discovered via enriched workflow logs.
2. Formalization of an extended event-log meta-model integrating performatives, actors, roles, units, and document flows.
3. Implementation of a modular prototype supporting detection of organizational forms and interaction protocols, with AGR (Agent–Group–Role) visualization.

Limitations include the need for full implementation of performance analytics and log conversion to XML, validation on larger real-world data sets, and formal evaluation against engineered baselines (Abdelkafi et al., 2012). A plausible implication is that future work may incorporate integrated mining of process, informational, and organizational perspectives, further advancing the field of workflow mining through multi-perspective analysis.

---

Reference: "DiscopFlow: A new Tool for Discovering Organizational Structures and Interaction Protocols in WorkFlow" (Abdelkafi et al., 2012)