Mise en place d’une solution Deep Learning pour la fouille de l’opinion exprimée dans les retombées de presse au profit des clients d’Imperium Media
Abstract #
Among the fields of application of data analysis for value creation, Data as a science is a
very rich domain that allows companies to structure and automate the processing of their data,
using a “Cloud” network which ensures the confidentiality of the information communicated.
This dissertation is part of the final project to obtain the degree of State Engineer at the
School of Information Sciences (ESI), which took place within Imperium Media, one of the
most recognized data providers in Morocco, for a four-month time interval.
The goal of this research is to develop a solution for automating the analysis of opinions
based on the impact of the press by using deep learning technology.
Achieving this goal requires three phases:
- To have a wide visibility vis-à-vis the existing within the company, its know-how and
the technologies used, in order to adapt the solution to the existing needs.
- Gather a lot of documentation about the key terms that make up the subject, namely
the analysis of feelings, deep learning and of course the relationship between the two.
- Design and implement the press releases analysis solution to derive the tone of each
article using convolutional neural networks (CNNs) and recurrent neural networks
(RNNs), and making use of important libraries offered by Python, such as TensorFlow,
Keras, NumPy and PyMongo libraries …
Thus, in order to chart the path towards the achievement of these objectives, we adopted a
direct observation method, a method of documentary research and field investigation.
This approach has led to a better design of the solution that responds perfectly to the
problem identified. Thus, we have been able to implement highly developed tools and very
efficient methods for storing, processing and analyzing large amounts of data using different
algorithms of Deep Learning.
Keywords: Deep Learning, opinion analysis, press releases, CNN convolutional neural
networks, recurrent neuron networks RNN, MongoDB, PyMongo, TensorFlow, NumPy,
Keras.