The PREDHIC project (Predicting readmission and heart failure mortality using natural language processing) is a 42-month ANR project between two computer science laboratories (LISN and LS2N) and two hospitals (Groupe Hospitalier Paris Saint-Joseph and CHU de Lille). As part of this project, we are exploring research directions associated with neural methods for text representation and classification, and evaluating their impact on a real-life task and data, in a multi-center setting: estimating the risk of re-hospitalization and death in heart failure patients, a question of great clinical interest. 

The project brings together specialists in automatic language processing from two computer science teams and specialists in medical information and heart failure from two hospitals. The aim of the post-doc is to design and evaluate classifiers for risk prediction based on both structured data (e.g. biology tests) and text from patients' medical records. 

The aim is to evaluate different strategies for capitalizing on the richness of these data, using early or late fusion approaches. The data set is already available and pre-processed. The variables of interest associated with the structured data have been identified and are the subject of an initial model based on logistic regression. Textual data are already annotated into entities of interest, based on an annotation scheme developed jointly by NLP specialists and cardiology clinicians. 

Skills/Qualifications

General, theoretical or disciplinary knowledge : Natural Language Processing

Operational skills : Learning with neural models

Soft skills : Rigorous, autonomous, team player

Specific Requirements

Computer science thesis; Interest in automatic language processing; Python language ; Github; Experience in learning with neural networks..

Education and/or qualification :   PhD Previous experience welcome for the position