Aubrey J. Rembert

Towards efficient business process clustering and retrieval: combining language modeling and structure matching

Large organizations tend to have hundreds of business processes. Discovering and understanding similarities among business processes can be useful to organizations for a number of reasons including better overall process management and maintenance. In this paper we present a novel and efficient approach to cluster and retrieve business processes. A given set of business processes are clustered based on their underlying topic, structure and semantic similarities. In addition, given a query business process, top k most similar processes are retrieved based on clustering results. In this work, we bring together two not wellconnected schools of work: statistical language modeling and structure matching and combine them in a novel way. Our approach takes into account both high-level topic information that can be collected from process description documents and keywords as well as detailed structural features such as process control flows in finding similarities among business processes. This ability to work with processes that may not always have formal control flows is particularly useful in dealing with real-world business processes which are not always described formally. We developed a system to implement our approach and evaluated it on several collections of industry best practice processes and real-world business processes at a large IT service company that are described at varied levels of formalisms. Our experimental results reveal that the combined language modeling and structure matching based retrieval outperforms structure-matching-only techniques in both mean average precision and running time measures.

Beyond workflow mining

In the domain of Business Process Management and Workflow Management Systems, the log of work transactions executed has been found to be a useful artifact. The ideas, work, and literature on workflow mining have been primarily concerned with examining the workflow event log to rediscover control flow. Workflow mining has generally been defined as “the process of extracting a workflow model from a log of executions of activities”. In fact, most of the literature specifically and narrowly is concerned with rediscovering the precedence relations amongst activities. It is generally a hidden assumption that all activities are known a priori because they are listed by label in the workflow event log. In this position paper, we explore the possibility of removing this assumption, and thus performing workflow discovery rather than precedence rediscovery. Workflow discovery does not assume that process structure or even activities are known a priori and is concerned with discovering a wholistic perspective of workflow. Workflow management systems are people systems that must be designed, deployed, and understood within their social and organizational contexts. Thus, we argue in this document that there is a need to expand the concept of workflow mining beyond the behavioral perspective to encompass the social, organizational, and activity assignment perspectives; as well as other perspectives. To this end, we introduce a general framework and meta-model for workflow discovery, and show one approach to workflow discovery in a multidimensional perspective.

Comprehensive workflow mining

Workflow Management Systems (WFMS) assist with the execution, monitoring and management of a process. These systems, as they are executing, keep a record of who does what and when (e.g. a workflow event log). The activity of using computer software to examine these records to produce useful knowledge about a process is called workflow mining. Currently, workflow mining research is narrowly focused on the rediscovery of control flow models. In this position paper, we present comprehensive workflow mining to broaden the scope of workflow mining. We present the concepts of comprehensive workflow mining using the Information Control Net (ICN) workflow modeling language.

Investigations on Stochastic Information Control Nets

Information Control Nets have been well used as a model for knowledge mining, discovery, and delivery to increase organizational intelligence. In this document, we extend the notions of classic Information Control Nets [15] to define new concepts of Stochastic Information Control Nets. We introduce a simple and useful AND-probability semantic and show how this probabilistic mathematical model can be used to generate probabilistic languages. The notion of a probabilistic language is introduced as a normalizer for comparisons of organizational knowledge repositories to organizational models. We discuss model-log conformance and present a definition of fidelity of a model. We show how to manipulate the residual error factor of this model. We describe a set of recursive functions and algorithms for generation of probabilistic languages from stochastic ICNs. We prove an important aspect of our generation algorithms: they generate probabilistic languages that are normalized. Since ICN models with loops generate infinitely many execution sequences, we present new notions of most probable sequence generation, and @e-equivalent approximation languages. These definitions can be applied to many aspects of organizational modeling including the process, the informational, and the resource perspectives. The model that we introduce here can be used to augment and expand on analyses that have been useful and insightful within varied enterprise information systems modeling and organizational analysis applications.

From informal process diagrams to formal process models

Process modeling is an important activity in business transformation projects. Free-form diagramming tools, such as PowerPoint and Visio, are the preferred tools for creating process models. However, the designs created using such tools are informal sketches, which are not amenable to automated analysis. Formal models, although desirable, are rarely created (during early design) because of the usability problems associated with formal-modeling tools. In this paper, we present an approach for automatically inferring formal process models from informal business process diagrams, so that the strengths of both types of tools can be leveraged. We discuss different sources of structural and semantic ambiguities, commonly present in informal diagrams, which pose challenges for automated inference. Our approach consists of two phases. First, it performs structural inference to identify the set of nodes and edges that constitute a process model. Then, it performs semantic interpretation, using a classifier that mimics human reasoning to associate modeling semantics with the nodes and edges. We discuss both supervised and unsupervised techniques for training such a classifier. Finally, we report results of empirical studies, conducted using flow diagrams from real projects, which illustrate the effectiveness of our approach.

An initial approach to mining multiple perspectives of a business process

In this work, we present some preliminary ideas about the development of an approach to mining different perspectives of a business process. Process mining, to date, has been narrowly concerned with mining the control-flow of a business process. There are very few process mining algorithms aimed at mining different business process perspectives. We believe one of the primary reasons for the paucity of process mining algorithms in perspectives other than control-flow is that there has been no general definition of what a business process perspective is. With this work, we provide a formal and general definition of a business process perspective, and present an approach to mine other business process perspectives using this definition.

Process Discovery Using Prior Knowledge

In this paper, we describe a process discovery algorithm that leverages prior knowledge and process execution data to learn a control-flow model. Most process discovery algorithms are not able to exploit prior knowledge supplied by a domain expert. Our algorithm incorporates prior knowledge using ideas from Bayesian statistics. We demonstrate that our algorithm is able to recover a control-flow model in the presence of noisy process execution data, and uncertain prior knowledge.

Discovering Shared Services from Cross-Organizational Software Specifications

During a standard software development process, organizations create text-based documents that describe software requirements, design, and implementation. These text-based specifications describe the functionality of future applications as they relate to an existing IT infrastructure. We suggest that these documents also implicitly describe core underlying service-based capabilities of the organization. In this paper, we describe an approach that (when provided with software specifications from multiple organizations) can recommend services shared by the multiple organizations represented. These approaches leverage the syntactic similarity of the specification text and semantic information as inferred from WordNet. Experiments show the effectiveness of this approach when processing real software requirements specifications in operational environments.

Learning the Control-Flow of a Business Process Using ICN-Based Process Models

In this paper, we present a process mining algorithm that discovers Activity Precedence Graphs (APG), which are control-flow models in the Generalized Information Control Net (ICN) family of models. Unlike many other control-flow models discovered by process mining algorithms, APGs can be integrated with other business process perspectives.

On Predicting Impacts of Customizations to Standard Business Processes

Adopting standard business processes and then customizing them to suit specific business requirements is a common business practice. However, often, organizations don’t fully know the impact of their customizations until after processes are implemented. In this paper, we present an algorithm for predicting the impact of customizations made to standard business processes by leveraging a repository of similar customizations made to the same standard processes. For a customized process whose impact needs to be predicted, similar impact trees are located in a repository using the notion of impact nodes. The algorithm returns a ranked list of impacts predicted for the customizations.