Construction et l’analyse d’un graphe de connaissance biomédicale en utilisant les techniques de Deep learning

Abstract #

Data science is a multidisciplinary field that uses knowledge from the fields of mathematics,

statistics and technology to study and evaluate data. It is a subject that deals with the identification,

representation and extraction of meaningful information from data sources in order to use them for

commercial or other purposes.

In this context, the objective of this thesis is to propose a generic approach for the construction of a

knowledge graph using Deep Learning techniques, through textual data in the format of clinical notes

of people in order to analyze them.

To meet this objective, we built a system that would automatically structure this data in a format that

would allow doctors and patients to quickly find the information they need. Specifically, we built a

named entity recognition model that would recognize entities such as drug, strength, duration,

frequency, adverse drug reactions, reason for taking the drug, route of administration, and form. In

addition, the model would also recognize the relationship between the drug and any other named entity.

This would allow healthcare professionals to examine not only individual entities, but all relationships

between them. It would also allow healthcare professionals to easily find relationships between a drug

and clinical notes, so that these drugs can be closely monitored.

And to showcase the work that has been done, we have set up a dissemination platform to make it easy

to visualize the prediction results.

Keywords: Deep Learning, knowledge graph, clinical note, BERT, HER, Semantic search, Neo4j,

graph