Process Knowledge Extraction and Knowledge Graph Construction Through Prompting: A Quantitative Analysis

The automated construction of process knowledge graphs from process description documents is a challenging research area. Here, the lack of massive annotated data, as well as raw text repositories describing real-world process documents, makes it extremely difficult to adopt deep learning approaches to perform this transformation. Indeed, the main challenge is to extract conceptual elements representing the actual entities or relations of the process model described within its corresponding natural language document. Large Language Models (LLMs) have shown promising results in supporting the extraction of structured knowledge from unstructured texts. Although several works explored this strategy to build or complete knowledge graphs, the exploitation of LLMs toward domain-specific knowledge base construction from scratch has not yet been investigated deeply. Our aim is to exploit the LLM capabilities to extract process knowledge from unseen natural language descriptions. In this work, we present a prompt-based in-context learning strategy to extract, from process descriptions, conceptual information that can be converted into their equivalent knowledge graphs. Such a strategy is performed in a multi-turn dialog fashion. We validate the accuracy of the proposed approach from a quantitative perspective. The results highlight the feasibility of the proposed approach within our low-resource scenarios and open interesting perspectives for future activities.