Authors:
Anas El Ouardi
;
Maryem Rhanoui
;
Anissa Benlarabi
and
Bouchra El Asri
Affiliation:
IMS Team, ADMIR Laboratory, Rabat IT Center, ENSIAS, Mohammed V University, Rabat, Morocco
Keyword(s):
Text Classification, Meta-learning, Few-shot Learning.
Abstract:
Text classification is one of the most prolific domains in machine learning. Present in a raw format all around us in our daily life Starting from human to human communication mainly by the social networks apps, arriving at the human-machine interaction especially with chatbots, text is a rich source of information. However, despite the remarkable performances that deep learning achieves in this field, the cost in therm of the amount of data needed to train this model still considerably high, adding to that the need of retraining this model to learn every new task. Nevertheless, a new sub-field of machine learning has emerged, named meta-learning it targets the overcoming of those limitations, widely used for image-related tasks, it can also bring solutions to tasks associated with text. Starting from this perspective we proposed a hybrid architecture based on well- known prototypical networks consisting of adapting this model to text classification and augmenting it with a non-linea
r classifier.
(More)