Tree-LSTM and Cross-Corpus Training for Extracting Biomedical Relationships from Text

Joël Legrand (1) , Yannick Toussaint (1) , Chedy Raïssi (1) , Adrien Coulet (1)

1 ORPAILLEUR - Knowledge representation, reasonning

A key aspect of machine learning-based relationship extraction algorithms is the availability of training data. Manually annotated corpora are valuable resources for this task, but the time and expertise required for their development explain that still few corpora are available. For tasks related to precision medicine, most of them are rather small (i.e., hundreds of sentences) or focus on specialized relationships (e.g., drug-drug interactions) that rarely fit what one wants to extract. In this paper, we experiment Tree-LSTM, to extract relationships from biomedical texts with high performance. In addition we show that relatively large corpora, even when focusing on a distinct type of relationships, can be use to improve the performance of deep learning-based system for relationship extraction tasks for which initial resources are scarce.

Mots clés: relationship extraction; pharmacogenomic relationships; Cross-corpus training; Deep learning; NLP; Tree-LSTM

Domaines: Apprentissage [cs.LG]; Bio-informatique [q-bio.QM]

Joël Legrand, Yannick Toussaint, Chedy Raïssi, Adrien Coulet. Tree-LSTM and Cross-Corpus Training for Extracting Biomedical Relationships from Text. DLPM2017 Workshop - 2nd International Workshop on Deep Learning for Precision Medicine, Sep 2017, Skopje, Macedonia. ⟨hal-01592802⟩