A Novel Deep Learning Multi-Modal Sentiment Analysis Model for English and Egyptian Arabic Dialects Using Audio and Text.

As emotions play an important role in human interaction, the need for sentiment analysis has become crucial for human-computer interaction. This paper proposes a new model named Audio-Text Fusion (ATFusion), for sentiment analysis that utilizes text and speech data to detect emotions. The model comprises local classifiers for audio and text inputs, followed by fusion using Group Gated Fusion (GGF) technique. Convolutional neural network (CNN), long short-term memory (LSTM) neural network and transformers are employed as building blocks for the local classifiers. The evaluation of the ATFusion model performance is demonstrated through experiments over the IEMOCAP dataset for the English language and the EYASE dataset for the Egyptian Arabic Dialect. The performance of the ATFusion Model compared to other state-of-the-art models achieved results of 76.213%, 75.146%, and 70.79%, 70.42%, unweighted accuracy and weighted accuracy over IEMOCAP and EYASE, respectively.