A Three-stage Neuro-symbolic Recommendation Pipeline for Cultural Heritage Knowledge Graphs
Krzysztof Kutt, Luiz do Valle Miranda, Elżbieta Sroka, Oleksandra Ishchuk
arXiv (Cornell University)
Problems Identified (4)
Cultural heritage semantic recommendation: Digital cultural heritage resources need advanced recommendation methods that can interpret semantic relationships among heterogeneous data entities.
Sparse heterogeneous metadata: Recommendation over the target cultural heritage graph is challenged by sparse and heterogeneous metadata.
Cultural heritage semantic recommendation: Digital cultural heritage resources need advanced recommendation methods that can interpret semantic relationships among heterogeneous data entities.
Sparse heterogeneous metadata: Recommendation over the target cultural heritage graph is challenged by sparse and heterogeneous metadata.
Proposed Solutions (5)
Three-stage neuro-symbolic recommender: The paper proposes a three-stage hybrid recommendation pipeline combining neural knowledge-graph methods with symbolic semantic filtering.
KG embedding ANN SPARQL pipeline: The methodology integrates knowledge-graph embeddings, approximate nearest-neighbour search, and SPARQL-driven semantic filtering for recommendations.
Embedding family evaluation and tuning: The work evaluates multiple knowledge-graph embedding families and performs hyperparameter selection for ComplEx and HNSW.
Three-stage neuro-symbolic recommender: The paper proposes a three-stage hybrid recommendation pipeline combining neural knowledge-graph methods with symbolic semantic filtering.
KG embedding ANN SPARQL pipeline: The methodology integrates knowledge-graph embeddings, approximate nearest-neighbour search, and SPARQL-driven semantic filtering for recommendations.
Results (3)
JUHMP knowledge graph evaluation:
Useful explainable recommendations:
Expert evaluation support:
Research Domain
Cultural heritage knowledge graph recommendation systems