The project will study the extraction of relation in medical texts in different settings:

• Recognizing relations between medical entities in EHRs. These relations will give connections between concepts in the semantic graph relative to each patient. The relations will be extracted applying NLP over the semantic graph for each patient. Examples of relations are "a sign appearing in a body part indicates a diagnosis", "a diagnosis is dealt with a treatment", or "a test will give a result".
• Relation extraction techniques will be proposed to recommend ICD-10 codes for the automatic annotation of medical records.
• Detecting pairs (drug, disease) that produce adverse effects, using supervised and unsupervised methods.