emotion detection

The ParsiAzma[1] challenges in 2023 focused on Improving Persian text analysis in social media. We designed four shared tasks: stance detection, sentiment analysis, emotion detection, and claim detection in social media posts. The goal of these challenges was to bring together various teams to develop the best models for these challenges and to establish a standard test platform for future Persian language research. A total of 28 teams participated, competing to solve the specified tasks. The most effective models in all shared tasks utilized the BERT model. Text embedding was first obtained using a BERT[2]-based model, followed by final predictions with either an MLP[3] or CNN[4]. Additionally, several meta-classifiers were developed as fusion models to leverage the strengths of individual models. The best results based on accuracy criteria for the four challenges—stance detection, sentiment analysis, emotion recognition, and claim detection—were 0.67, 0.67, 0.45, and 0.56, respectively. These results indicate that emotion detection has lower accuracy than the other three tasks, highlighting its complexity.

Text analysis has been one of the issues in recent research to identify users' sentiments. Most studies have identified sentiments' positive and negative polarity in Persian, and limited research has been done on analyzing emotions in Persian sentences by covering the primary emotional states. In this study, first, a dataset of emotional sentences was prepared to label six basic emotional states, JAMFA. This dataset contains 2350 sentences and (31222 words). This paper presents two models, efficient BERT-BiLSTM(EBB) and XLM-R Catboost(XLM-RC), that enhance the performance of the Persian text emotion classification. This study has the advantages of human intelligence methods and statistical approaches to achieve better accuracy in sentence labeling. The evaluation indicates the accuracy of labeling is 92%, and the reliability of the dataset based on the type of emotions is 88%. The results show that the models at best achieved 86\% accuracy in basic emotion classification and an 81% F-score in binary classification.

In the modern age, written sources are rapidly increasing. A growing number of these data are related to the texts containing the feelings and opinions of the users. Thus, reviewing and analyzing of emotional texts have received a particular attention in recent years. A System which is based on combination of cognitive features and deep neural network, Gated Recurrent Unit has been proposed in this paper. Five basic emotions used in this approach are: anger, happiness, sadness, surprise and fear. A total of 23,000 Persian documents by the average length of 24 have been labeled for this research. Emotional constructions, emotional keywords, and emotional POS are the basic cognitive features used in this approach. On the other hand, after preprocessing the texts, words of normalized text have been embedded by Word2Vec technique. Then, a deep learning approach has been done based on this embedded data. Finally, classification algorithms such as Naïve Bayes, decision tree, and support vector machines were used to classify emotions based on concatenation of defined cognitive features, and deep learning features. 10-fold cross validation has been used to evaluate the performance of the proposed system. Experimental results show the proposed system achieved the accuracy of 97%. Result of proposed system shows the improvement of several percent’s in comparison by other results achieved GRU and cognitive features in isolation. At the end, studying other statistical features and improving these cognitive features in more details can affect the results.