Verifying Resource Compliance Requirements from Natural Language Text over Event Logs

Process compliance aims to ensure that processes adhere to requirements imposed by natural language texts such as regulatory documents. Existing approaches assume that requirements are available in a formalized manner using, e.g., linear temporal logic, leaving the question open of how to automatically extract and formalize them for verification. Especially with the constantly growing amount of regulatory documents and their frequent updates, it can be preferable to provide an approach that enables the verification of processes with requirements in natural language text instead of formalized requirements. To this end, this paper presents an approach that copes with the verification of resource compliance requirements, e.g., which resource shall perform which activity, in natural language over event logs. The approach relies on a comprehensive literature analysis to identify resource compliance patterns. It then contrasts these patterns with resource patterns reflecting the process perspective, while considering the natural language perspective. We combine the state-of-the-art GPT-4 technology for pre-processing the natural language text with a customized compliance verification component to identify and verify resource compliance requirements. Thereby, the approach distinguishes different resource patterns including multiple organizational perspectives. The approach is evaluated based on a set of well-established process descriptions and synthesized event logs generated by a process execution engine as well as the BPIC 2020 dataset.