This project aims to develop advanced (deep) machine learning techniques for information extraction from knowledge graphs in the context of cultural heritage. A knowledge graph for storing and managing heterogeneous data (images, video, audio, text) will be implemented using established graph database software, for example, with Neo4J. Then the research activity will focus on the development of innovative methods for static and dynamic clustering on top of the knowledge graphs. To this end, groups of similar nodes will be extracted considering both constrained and unconstrained scenarios. In this context, graph neural networks will play a central role in generating highly expressive latent representations of knowledge graph nodes. Such embedding will be used to perform different actions on the knowledge graph (clustering, link-prediction, node classification) with the goal of select and/or automatically generate custom content for users.
Plus d’informations :
[Télécharger l'annonce - PDF 240 Ko]