Academic stories: Pnina Sofferacademic stories
Narrated by Pnina Soffer
My name is Pnina Soffer, I work at the University of Haifa in Israel. I belong to the Department of Information Systems, which is a part of the Faculty of Social Sciences in the university. This untraditional setting facilitates a lot of inter-disciplinary work, where technology is developed for humans, in collaboration with researchers of diverse areas, including psychology, economics, healthcare, humanities (history, archaeology), and many others. My department has 14 faculty members and around 450 students – mostly undergraduates but also master and doctoral students. It is a very friendly department with a family-like atmosphere and a lot of collaboration among the researchers.
I have done research on business process management since my PhD thesis (which I completed in 2002 at Technion). At first, I was mostly dealing with process modeling. Two main research streams I had then were theoretical ontology-based foundations of process models and empirical investigations of process modeling. I was trying to tie the theoretical foundations with possible manifestations at the cognitive processes of process modeling and understanding. At that time, process mining research was mainly focused on control flow, and so I was not involved in this area, since my interest was in additional aspects of process models. I was particularly interested in the goal concept in relation to processes, and so in 2007 this interest has motivated the PhD work of Johny Ghattas, which aim was to learn from past experience how to improve the achievement of goals in processes. Johny’s PhD, co-supervised by me and Mor Peleg, was my first engagement in process mining, and was in fact one of the earliest prediction works in this area. The main vision was to extend process mining beyond its control-flow orientation, and to augment the control flow by contextual data for predicting business outcomes. With these predictions, specific operational recommendations could be given, considering the specific context of the process, for achieving the best business outcomes that could be predicted. Since then, many research efforts have been devoted to prediction and much progress has been made. I continued working on process mining projects, and my work was, and is still, devoted to additional aspects besides the control flow perspective.
I view process mining as an amazing area, with many strong researchers who put much effort into developing algorithms, constantly extending and improving the capabilities of process mining tools. Yet, I don’t consider myself as one of the hard-core algorithm developers. What I find fascinating is how process mining can be used and how it can be extended when addressing more general problems related to BPM and to human behavior.
One example of the kind of research I do is a project investigating workarounds in business processes and attempting to address their underlying reasons as leverage for process improvement. The idea is that if we understand the motivation for performing workarounds, we can design solutions to real problems that exist in the process and motivate workarounds. This project takes a very broad view, including qualitative studies and interviews of users in organizations, behavioral theories, and process improvement methods. Process mining is used as a quantitative tool for detecting and measuring workarounds and their business impacts. A main challenge is to establish a mapping between what we know about business situations and human behavior and patterns that can be detected in an event log. Detecting such patterns can rely on existing tools and algorithms, but may also imply a need for new algorithms to be developed. An example we came across in this study is a common pattern of “splitting orders” – where orders (or other objects) are split and handled in two or more cases to avoid a need for additional controls, required over a certain threshold (e.g., manager’s approval is required for orders above 5000 Euro). Addressing each case separately (e.g., checking its conformance) would yield normative and conformant behavior, and would not reveal this workaround. Hence, a need was identified – to develop a method for detecting cross-case patterns of this kind. Very little work has been done so far on cross-case patterns, and most of the process mining methods capture and analyze behavior within each case. I personally consider the case-driven view to be one of the strengths of process mining, distinguishing process mining from other kinds of data analysis. Yet, I believe that the case view should be complemented by addressing cross-case patterns of behavior. This view is relevant for discovery, analysis, conformance checking (against cross-case constraints), and prediction (using inter-case properties). And so this is a current direction in my process mining research. In parallel, I also deal with cross-case patterns from a formal model-based perspective.
Another topic I am interested in is a holistic view of process mining as a process. I came to this when the PhD student Arava Tsoury (co-supervised by me and Iris Reinhartz-Berger) became very frustrated with the partial availability of data attributes in event logs (or with a need to pre-specify what subset of data should be included in a log). This limited availability of data hindered comprehensive investigations of data impact over process behavior she was intending to perform. Attempting to overcome this, she turned to additional data sources, such as the transaction (redo) log of a database as well as the database itself, but then challenges related to granularity level of the data were faced, since database operations (insert/update/delete data items) are at a much finer granularity level than event logs of processes. She then started to explore the possibilities arising from a combination of these data sources, supporting combinations to be formed in an ad-hoc manner for specific analyses based on an underlying established mapping. A main and basic idea is that the exploration steps in the process of process mining can vary, and can be determined on-the-fly. Furthermore, different types of information and granularity levels may be required at different steps of this process. In a preliminary study, we let students use a platform which supports ad-hoc combinations of an event log, a database transaction log, and database tables. Using simulated data, we asked the students high-level business questions (e.g., what could be the reasons for delays in the process). We logged their interactions with the platform as they were working to be able to track the way they approach these questions, and realized this is quite an unexplored territory…
While many of the challenges along the process of process mining have long been recognized and discussed already in the process mining manifesto, most of the research efforts so far related to specific tasks, addressing each one separately. Even advanced tools, which allow the user to “play” and explore the analysis results, and to employ different visualizations, are not (as far as I know) based on extensive user studies and theoretical grounds. I believe we need to establish a profound understanding of how analysts think along the overall process of process mining, starting from high-level business questions, selecting data sources, inspecting, cleansing, and preprocessing the data, selecting and applying process mining algorithms, and finally, interpreting and presenting the results. Understanding this process from a human cognition perspective would illuminate the needs that a holistic support of this overall process should satisfy.
So this is a line of research I intend to pursue. I intend to do that in collaboration with colleagues with whom I had a good collaboration in the past, investigating the process of process modeling, Barbara Weber and Andrea Burattin. Hopefully, additional researchers will join, and together we will be able to do new and exciting work and contribute new perspectives to process mining research.
For those who are interested to read more about the works that I mentioned, here are some links:
- Papers related to the PhD thesis of Johny Ghattas: https://is-web.hevra.haifa.ac.il/staff/spnina_files/publications/JDSS2013.pdf and http://is.haifa.ac.il/~morpeleg/pubs/BPM2009_Prohealth2009.pdf
- An early paper about using process mining for workaround detection: https://is-web.hevra.haifa.ac.il/staff/spnina_files/publications/JSosym2014Workarounds.pdf
- A (non process mining) paper from the workarounds project: https://www.researchgate.net/publication/340771954_Workarounds_in_Business_Processes_a_Goal-based_Analysis
- Arava Tsoury’s use of a combination of event log with databases and transaction logs: https://link.springer.com/chapter/10.1007/978-3-030-00847-5_6 and https://www.researchgate.net/publication/338346667_Impact-Aware_Conformance_Checking
- Academic stories: Pnina Soffer
- From Research Vision to Startup: The Story of Apromore
- Exploring newland: Blockchains and process mining
- BPI Challenges: 10 years of real-life datasets
- End-user’s corner: Bruna Pereira
- ICPM 2020 goes fully virtual!
- Special Issue on Process Mining and Emerging Applications (MDPI Algorithms journal)
- This article has been updated on May 14 2020, 19:42.
- Narrated by Pnina Soffer