Learning the users' emotions is essential in many applications, such as social robots used as communication assistance for education and entertainment [9] This article describes an acceptable and natural interaction between Social Robots and children. The thermal facial reaction of youngsters, i.e., the nose tip temperature signal, was recorded and classified in real-time using the Mio Amico Robot during an experimental session. The categorization was performed by comparing the thermal signal analysis-classified emotional state to the emotional state recorded by Face reader 7. An empathic robot in [7] to recognize human emotions through facial expressions and automatically respond to these specific emotional states produced a state-of-the-art accuracy rate of 95.58%. Using Convolutional Neural Network CNN and a bank of Gabor filters in different experiments for feature representation and employs SVMs and MLPs as classifiers.
Customer feedback detection, as in [13] where a multimodal affect recognition system developed to classify whether a customer likes or dislikes a product examined at a counter, by analyzing the consumer's facial expression, hand gestures, body posture, and voice after testing the product. Hand gesture recognition is a component of Human action recognition (HAR) and is widely used in scientific research; it is critical for interacting with deaf individuals. [10] proposes an approach divided into transfer learning through Alex Nets and hyperparameter tuning through ABC, GA, and PSO algorithms. The methodology produced effective outcomes with an average accuracy of 98.09 percent, beating the best work in the medical sector in [21]; computational analysis techniques are used to measure the emotional facial expression of people who have Parkinson's disease (PD). Since PD experiences hypomania, which often causes a reduction in facial expression, it is important to examine an experimental pilot work for masked face detection in PD. This experiment achieved an accuracy of 85% on the testing images using a deep learning-based model.
A novel methodology for incorporating human emotion into intelligent computer systems is presented [14]. It has been proposed as a method to elicit emotional information from users. A hybrid cloud intelligence model using an adaptive fuzzy method with a high degree of interpretability achieves a satisfactory performance accuracy of 81.39% using Facebook's sentiment analysis API.
The use of emotion lexicons is of great importance in the emotion classification task. Many lexicons in different languages have been built as [16],[22], [23], [19], [24], [25], [26] in English, Polish and French. In Arabic, many efforts have been recently made in building emotion lexicons as ArSEL(Arabic Sentiment
and Emotion Lexicon)[18]., ArSEL has been constructed automatically by using three lexical resources: epecheMood, English WordNet and ArSenL. These lexicons helped in improving the sentiment classification model accuracy.
Multilabel emotion classification is a hot topic in emotion analysis tasks since it represents real-life situations where the human may express a mixture of emotions in his text simultaneously. For example, the text may express happiness, love, optimism, or maybe sadness and pessimism, so it is more beneficial to build such models with more than one output emotion for each input text. The following few paragraphs below browse some recent efforts in Arabic multilabel emotion classification:
EMA (Emotion Mining in Arabic) [16] performs Emotion and Sentiment mining on Arabic tweets. First, preprocessing steps are performed, first applying normalization rules adopted by [27], including removing diacritics or taskeel and hamza removal, then removing elongations and non-Arabic letters. Next, most frequent emojis have been replaced with the corresponding Arabic word using a manually created lexicon to replace each emoji. Finally, using ARLSTEM [28] for stemming. Then in the feature selection stage, the author tried different features separately, but the word embedding from AraVec proved to be the best feature. The tweet is finally classified either as neutral or as one or more of 11 emotions (anger, disgust, anticipation, joy, love, optimism, fear, pessimism, sadness, trust, surprise). Linear SVC performed best among all classifiers tested, with a test accuracy of 0.489.
TW-Star [29] uses different preprocessing stemming (Stem), lemmatization (Lem), stop words removal (Stop), and common emoji recognition (Emo). The preprocessed tweets are then classified by a multilabel classifier based on Binary Relevance (BR) using the Support Vector Machines (SVM) with Term Frequency Inverse Document Frequency (TF-IDF) features. Several experiments with different combinations of preprocessing achieved the best results of accuracy 0.465 using a combined of (Emo + Stem + Stop).
TeamUNNC [30] performs tokenization, removal of white spaces, and treating punctuations as individual words. In the second stage, word2vec embedding AraVec [20] combined with Affective Tweets Weka-package features. Finally, classification is implemented with a fully connected neural network having three dense hidden layers and an SGD (Stochastic Gradient Descent) optimizer. The model achieved an accuracy of 0.446, exceeding the baseline model accuracy.
In [31], feature vectors developed using the Doc2Vec model; then, the Random Forest RF algorithm was used for classification; Doc2Vec size varied from 10 to 1000 with an incremental of 10 iterations. The number of decision trees used in the forest ranged from 10 to 150, with an incremental of 10 in each iteration. The maximum tree depth in the algorithm varied from 2 to 20 and increment by 1 in each iteration. This model obtained an accuracy of 0.25.
TeamCEN [32] Uses Globe vector representation[33] for representing the words into vectors.; it depends on word-word co-occurrence statistics. Then the presentation of the tweet is made by using aggregated sum and dimensionality reduction of the glove vectors of the words in that tweet. The classification is then done using Random Forest RF and support vector machine SVM.
Our work analyzes a deep learning model for multilabel emotion classification in Arabic tweets described in the Proposed Method section below.