Deep Learning for Non Verbal Sentiment Analysis: Facial Emotional Expressions

EasyChair Preprint no. 3014

Abstract

People are of a lazy nature and always look for the easiest ways to express themselves and share their experiences and opinions. Due to the popularity of social networks, and to the images expressivity, people have the ability to express themselves throught their use. Our work is about non verbal sentiment analysis using one of the Deep Learning models: CNN (Convolutional Neural Networks). Specifically, we are interested in analyzing the sentiment expressed in facial expressions according to Kaggle's Dataset fer2013 for facial emotion recognition based on the emotions defined by the famous psychologist Ekman namely joy, anger, fear, disgust, sadness and surprise, neutrality is added to the six emotions. Thus, different proposed architectures are used and compared to determine the parameters that affect the results.

The best evaluation resulted in details of around 0,88 showing that the number of convolution layers, the batch_size, the dropout and the epoch number have an impact on the results. However using a CPU cost us a lot which proves that the use of a GPU when using huge amount of data is better and guarantee good results .

Keyphrases: Convolutional Neural Network (CNN), Deep Learning (DL), Emotionnal Facial Expression(EFE), image classification, Sentiment Analysis (SA)

@Booklet{EasyChair:3014,
  author = {Nour Meeki and Abdelmalek Amine and Mohamed Amine Boudia and Reda Mohamed Hamou},
  title = {Deep Learning for Non Verbal Sentiment Analysis: Facial Emotional Expressions},
  howpublished = {EasyChair Preprint no. 3014},

  year = {EasyChair, 2020}}