L’intégration de connaissances externes dans les modèles neuronaux est très étudiée pour améliorer les performances des modèles de langue pré-entraînés, notamment en domaine biomédical. Dans cet article, nous explorons la contribution de plongements de bases de connaissances à une tâche d’extraction de relations. Pour deux mentions d’entités candidates dans un texte, nous faisons l’hypothèse que la connaissance de relations entre elles, issue d’une base de connaissances (BC) externe, aide à prédire l’existence d’une relation dans le texte, y compris lorsque les relations de BC sont différentes de celles du texte. Notre approche consiste à calculer des plongements du graphe de BC et à estimer la possibilité pour chaque paire d’entité du texte qu’elle soit reliée par une relation de BC. Les expériences menées sur trois tâches d’extraction de relations en domaine biomédical montrent que notre méthode surpasse le modèle PubMedBERT de base et donne des performances comparables aux méthodes de l’état de l’art.

Entity normalization (or entity linking) is an important subtask of information extraction that links entity mentions in text to categories or concepts in a reference vocabulary. Machine learning based normalization methods have good adaptability as long as they have enough training data per reference with a sufficient quality. Distributional representations are commonly used because of their capacity to handle different expressions with similar meanings. However, in specific technical and scientific domains, the small amount of training data and the relatively small size of specialized corpora remain major challenges. Recently, the machine learning-based CONTES method has addressed these challenges for reference vocabularies that are ontologies, as is often the case in life sciences and biomedical domains. And yet, its performance is dependent on manually annotated corpus. Furthermore, like other machine learning based methods, parametrization remains tricky. We propose a new approach to address the scarcity of training data that extends the CONTES method by corpus selection, pre-processing and weak supervision strategies, which can yield high-performance results without any manually annotated examples. We also study which hyperparameters are most influential, with sometimes different patterns compared to previous work. The results show that our approach significantly improves accuracy and outperforms previous state-of-the-art algorithms.

This paper presents the fourth edition of the Bacteria Biotope task at BioNLP Open Shared Tasks 2019. The task focuses on the extraction of the locations and phenotypes of microorganisms from PubMed abstracts and full-text excerpts, and the characterization of these entities with respect to reference knowledge sources (NCBI taxonomy, OntoBiotope ontology). The task is motivated by the importance of the knowledge on biodiversity for fundamental research and applications in microbiology. The paper describes the different proposed subtasks, the corpus characteristics, and the challenge organization. We also provide an analysis of the results obtained by participants, and inspect the evolution of the results since the last edition in 2016.