A Generative AI application for Qualitative Automatic Population of Multimedia Knowledge Graphs

Recent advances in the field of artificial intelligence have created new opportunities for innovation in multimedia technologies. In particular, Knowledge Graphs (KGs), tools designed to manage knowledge efficiently and organise data by emphasizing relationships among entities, require high attention to the quality of the information they manage when applied to multimedia domains. Multimedia Knowledge Graphs pose additional challenges, as the curation and population of these structures often rely on manual quality assessment procedures, resulting in increased time and resource demands.In this paper, we propose an approach for the automatic population of a Multimedia Knowledge Graph, which employs generative AI techniques for image synthesis, introducing an evaluation strategy for assessing image quality. Our strategy computes a quality index from established evaluation metrics, substantially reducing routine checking time and bias while preserving expert judgment for unclear or safety-critical cases. Utilising our proposed methodology, we constructed a Multimedia Knowledge Graph from its textual counterpart. We then evaluated its effectiveness by training neural network models on the synthetic Multimedia Knowledge Graph. Quality index-based rankings track expert judgments, achieve faster evaluation times, match original training performance, and indicate scalable generalization. Our approach provides a reproducible alternative, minimizing manual procedures while maintaining quality standards.