期刊名称：International Journal of Advanced Computer Science and Applications(IJACSA)
印刷版ISSN：2158-107X
电子版ISSN：2156-5570
出版年度：2022
卷号：13
期号：4
DOI：10.14569/IJACSA.2022.01304112
语种：English
出版社：Science and Information Society (SAI)
摘要：Now-a-days people use social media websites for different activities such as business, entertainment, following the news, expressing their thoughts, feelings, and much more. This initiated a great interest in analyzing and mining such user-generated content. In this paper, the problem of emotion detection (ED) in Arabic text is investigated by proposing an ensemble deep learning approach to analyze user-generated text from Twitter, in terms of the emotional insights that reflect different feelings. The proposed model is based on three state-of-the-art deep learning models. Two models are special types of Recurrent Neural Networks RNNs (Bi-LSTM and Bi-GRU), and the third model is a pre-trained language model (PLM) based on BERT and it is called MARBERT transformer. The experiments were evaluated using the SemEval-2018-Task1-Ar-Ec dataset that was published in a multilabel classification task: Emotion Classification (EC) inside the SemEval-2018 competition. MARBERT PLM is compared to one of the most famous PLM for dealing with the Arabic language (AraBERT). Experiments proved that MARBERT achieved better results with an improvement of 4%, 2.7%, 4.2%, and 3.5% regarding Jaccard accuracy, recall, F1 macro, and F1 micro scores respectively. Moreover, the proposed ensemble model showed outperformance over the individual models (Bi-LSTM, Bi-GRU, and MARBERT). It also outperforms the most recent related work with an improvement ranging from 0.2% to 4.2% in accuracy, and from 5.3% to 23.3% in macro F1 score.
