Welcome to this project !
The topic covered here is Sentiment Analysis in texts written
in French language.
For that, we employ a Recurrent Neural Network that we build and run thru the Tensorflow / Keras framework.
The architecture of the model is based on dual bi-directionnal GRU cells
and it employs fastText word embeddings.
We train this model using tranfer learning from rated product reviews
that have been web-scrapped using the BeautifulSoup python library
(the web-scraping code is not provided,
but the collected data is).
The figure on the left shows the structure of this project.
There are two key points to notice :
- A dedicated custom python package named my_NLP_RNN_fr_lib has been developped to serve this project.
- There's a whole sub-section to the herein project, detailled separately, on hyperparameters optimization,.
It can be found there
.
Spoiler alert : we deal with random search first, then XGBoost + scikit-learn
are called to get an extra edge.
The French-Text Sentiment Analysis project we're dealing with here is explained in details and accompagnied with full running python code in a walkthrough Jupyter Notebook.
KEYWORDS :
Tensorflow, Keras,
GRU, RNN, NLP, fastText,
web-scraping, BeautifulSoup,
transfer learning, french sentiment analysis