GERoMe – A Novel Graph Extraction Robustness Measure

Abstract

The extraction of graph structures in Euclidean vector space is a topic of interest with applications in many fields, e.g., the biomedical domain. While a number of different approaches have been presented, a quantitative evaluation of those algorithms remains a challenging task: Manual generation of ground truth for real-world data is often time-consuming and error-prone, and while tools for generating synthetic datasets with corresponding ground truth exist, this data often does not reflect the complexity in morphology and topology that real-world scenarios show. As a complementary or even alternative approach, we propose GERoMe, a novel graph extraction robustness measure, which quantifies the stability of algorithms that extract multigraphs with associated node positions from non-graph structures. Our method takes edge-associated properties into consideration and does not necessarily require ground truth data. Moreover, available ground truth information can be incorporated to additionally evaluate the correctness of the graph extraction algorithm. We demonstrate the usefulness and applicability of our approach in an exemplary study on synthetic and real-world data.