Assessing feeling in textual data extracted from the Internet using a neural network and emojis
(A2019-9248)

AUTHORS

Alice Crépin, University of Paris Dauphine ; Paul Valentin Ngobo, University of Paris Dauphine/ PSL

KEYWORDS

affect; neural model; textual data

ABSTRACT

When a consumer expresses himself on the Internet, he can sometimes represent his affect by a rating (such as a number of stars for example). But how do we assess the expressed affect, when this type of notation is not available and the dataset is too important for human analysis? In this communication we present a method assessing the affect expressed in large textual datasets, by using emojis and a neural network model on Python. We apply this method to a practical example: textual data extracted from comments under YouTube videos. The dataset we use has been constructed for a doctoral work and is made up of more than two million comments collected on about two thousand and five hundred videos. The neural network model presented in this research obtains an average accuracy rate of 89% when used on our dataset. In this work we present the steps necessary to the construction of such a neural model and concrete results of its application.