Auto-Generating Business Process Model from Heterogeneous Documents: A Comprehensive Literature Survey

Automatically generated business processes can be sourced from documents in the form of structured text or natural language. Although there have been several studies discussing generating business processes, there is still no research using systematically reviews as the source documents. This study presents a systematic literature review on document sources, methods, and challenges in generating business processes. We conducted a systematic literature review published from 2017 to early 2022 and identified 24 main studies discussing the sources of documents in generating business processes. We formulated and applied inclusion and exclusion criteria in two stages to determine the most relevant studies for our research goal. This literature review found that the most frequently used document sources were textual business rules (structured rules), using case diagrams (14 main studies), event logs (7 studies), and natural language text (3 studies) including customer feedback. In the aspect of the method, the most widely used method is in the field is natural language processing (NLP), followed by other methods such as semantic knowledge engineering (SKE), fuzzy, graph-based, deep learning, and the combination of NLP with deep learning. Meanwhile, the challenges faced in generating business processes include text preprocessing or document extraction, integration of business rules with business processes, and challenges in the form of time interval constraints, activity sequences, dummy activities, or invisible tasks that are generally found in event logs.