Experiments with natural language queries on RDF vs. XML-serialized BPMN diagrams
Damaris Naomi Dolha, Robert Andrei Buchmann · Procedia Computer Science · 2024
Our study reports a comparative analysis of natural language interactions with BPMN models, specifically contrasting semantic graphs generated from diagrams by the RDF export of Bee-Up, against the traditional standard BPMN 2.0 XML export from standard-compliant tools (in our case, SAP Signavio Process Transformation Suite). Utilizing varied prompt engineering techniques, the study evaluates the efficacy of GPT-based services from OpenAI in interpreting the nuanced semantic network structures of BPMN-as-RDF and the standard control flow hierarchical decomposition of BPMN-as-XML. Although image-based multi-modal interpretation of BPMN diagrams is also available in such services, our work is motivated by the fact that most BPM systems deliver structured serializations and not images through their APIs; moreover, any data stored in diagrams cannot be scrutinized by computer vision capabilities, being set as annotations in most tools. By exploring both the challenges and effectiveness of utilizing natural language in interacting with BPMN models, the analysis underscores the ability of RDF to mediate semantic richness and open-ended extension of procedural knowledge compared to the closed-world of the XML interchange schemas. Diagrammatic environments are thus encouraged to pursue this as a potential convergence between different means of knowledge representation.