Le traitement du langage naturel pour l'analyse des données textuelles avec les méthodes machine-learning et deep-learning: Une étude de cas sur les demandes des clients de la Caisse Marocaine des Retraites.

Abstract #

**

Customer relations are one of the priorities of companies and organisations, the customer is king, so these

requests and complaints must be taken seriously. These requests can come from different sources, and

specifically through innovative technology platforms and tools. With the increase of these requests, the need to

analyse them arises. These requests, which are unstructured, express the need, dissatisfaction or others must be

analysed and studied to present a quality service with the required quality.

Generally, organisations rely mainly on structured data to detect and predict customer needs because of the

difficulty of detecting needs and desires from textual data. Hence the importance of natural language processing

(NLP) techniques.

This report summarises the work carried out, which aims to analyse the requests of the clients of the Moroccan

Pension Fund in order to improve the quality of service. This project is divided into three phases, namely the

extraction of themes present in the requests with the LDA algorithm, an analytical phase and a phase for the

development of an automatic response system. In general, this project aims to exploit natural language

processing, machine learning and deep learning techniques to build a model capable of analysing customer

requests.

In this sense, we carried out a Topic Modeling consisting of detecting the themes residing in the customer

requests that reached a Topic coherence value equal to 57%, then we conducted an analytical study according

to designed KPIs. Finally, we carried out an automatic response system with a deep learning model which

achieved a better accuracy of 94%.