Automated process weakness identification based on social media posts

• Applicant: Professor Dr. Henrik Leopold
• Subject Area: Data Management, Data-Intensive Systems, Computer Science Methods in Business Informatics
• Term: since 2021
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 456993549

Against the background of intense competition and rapidly changing demands from customers, many organizations strive for continuously improving their business processes. Over the last couple of years, process mining techniques have become increasingly popular to support this endeavor. The core idea of process mining techniques is to reconstruct how a process is executed by analyzing so-called event logs. These event logs are extracted from IT systems and, therefore, provide insights into what exactly happened in the organization. Among others, process mining techniques allow to detect undesired patterns, bottlenecks, and compliance issues. However, while the outcomes and insights produced by existing process mining techniques are valuable input for process improvement initiatives, they do not automatically identify concrete process weaknesses. This means that inferring what kind of problems exist in a process and where these exist still requires the extensive involvement of domain experts. Therefore, the goal of this proposal is to address the problem of manual work in the context of process improvement and develop a technique that automatically identifies process weaknesses relating to specific events. To achieve this, we exploit a widely available textual resource that captures the customer’s perspective: social media posts. The core idea is to identify process-related weaknesses described in social media posts and link them to the events of an event log. In this way, we are able to automatically determine what process weaknesses exist and to which events of the process they relate.To achieve this, three main challenges need to be overcome: 1) The relationship between the sentences from social media posts and event logs is partial. This means that only a few sentences of a considered post actually relate to a process weakness. 2) A single social media post may relate to multiple events from one or more event logs. 3) Both social media posts and event labels are problematic from a natural language analysis perspective. Event labels only contain a few words and may not represent proper sentences Social media posts are also rather short and often suffer from poor grammar and informal abbreviations. This project will address these challenges by defining a novel technique that combines natural language processing with optimization-based alignment creation. Furthermore, it will demonstrate the practical value of the technique by applying it to real-world data.

• DFG Programme: Research Grants
• International Connection: Netherlands
• Cooperation Partner: Professor Dr.-Ing. Hajo A. Reijers