A Best Paper Award for the BIRD team!
Congratulations to the BIRD team, winner of a Best Paper Award at EKAW2022 conference!
The EKAW conference (23rd International Conference on Knowledge Engineering and Knowledge Management) took place in Bolzano (Italy) from September 26 to 29. The conference focuses on all aspects of knowledge acquisition, modeling and management, whether theoretical, methodological, experimental or applicative.
This prize rewards the work of Nicolas Hubert, PhD student in the BIRD team at Loria and in the ERPI laboratory, Pierre Monnin (researcher at Orange, former PhD student in the Orpailleur team), Armelle Brun (professor in the BIRD team) and Davy Monticolo (professor in the ERPI laboratory) for their article “New Strategies for Learning Knowledge Graph Embeddings: The Recommendation Case”. In this work, the team presents machine learning models for recommendation and introduces Sem@k, a new semantic evaluation metric.
This paper addresses the problem of recommendation within knowledge graphs, approached as a link prediction task. Using machine learning models, the entities and relations of the graph are immersed in a low-dimensional vector space whose training requires the generation of false examples, in order to improve the representation induced by the discrimination between positive and negative examples. The proposed approach consists in pre-training a model on the whole graph, then refining it by considering only a reduced portion of the graph.
This work shows that this inexpensive specialization step is very advantageous to the quality of recommendations. It also addresses the limitations of traditional evaluation metrics from a semantic point of view, and proposes a new metric to rectify this: Sem@K, which provides a more exhaustive view of the quality of such models.
- Read the article: New Strategies for Learning Knowledge Graph Embeddings: The Recommendation Case
- More information about the BIRD team: bird.loria.fr
- Know more about the EKAW conference: ekaw2022