Posos data challenge

Main goal

Given a set of questions about diverse pharmacology problems, predict the sub-domain (category) of question in order to clarify the intent of the query. Formally, the problem is a text classification problem where texts are a set of phrase and the categories are 52 mutually exclusive intentions as designed by professional pharmacologists.

## Useful links:

Challenge link: Ecole Normale Supérieure challenge website

TODO

• extraction des mots les plus représentatifs d'une catégorie
• revoir le preprocess (notamment la correction orthographique)
• rafiner le modèle avec les mots extraits (matrice des cosine puis convolutions)
• nettoyer le workflow