Implementation of Dynamic Art Curation Engine in Global Art Collection Archive

This paper proposes an implementation method of a Dynamic Art Curation Engine within the Global Art Collection Archive. The manner in which information retrieval is handled in an integrated art archive is crucial, as it directly influences the dynamism and creativity of the art viewing experience. The Dynamic Art Curation Engine is an artwork retrieval system that leverages the pre-trained machine learning models of diverse art interpretations by museums and curators, providing a dynamic art experience by presenting artworks classified according to these interpretations. The curation engine is composed of four primary components: 1) a pretrained curation model developed from specific art interpretation; 2) curation model metadata generated by applying the curation model to artworks stored in the Integrated Art Archive; 3) dictionary metadata, which addresses data heterogeneity across museum collections; and 4) a relevance calculator that assesses relationships between artworks based upon their metadata. As a prototype, we developed a Dynamic Art Curation Engine that incorporates a pre-trained model based on the catalogs of three exhibitions and integrated it into the Global Art Collection Archive, which houses over one million art pieces from four museums. Our experiments addressed two main objectives: the curation of works by multiple artists based on specific exhibition categories, and the curation of works from arbitrary images. The results of these experiments confirm the effectiveness of the proposed method, thereby demonstrating its feasibility.