Deep Learning Techniques for Multimodal Sentiment Analysis

Abstract

Multi-modal sentiment analysis, as one of the most cutting-edge studies in artificial intelligence, has achieved multiple achievements and successes on account of the development of deep learning models. Due to the information explosion in social media, the data format is more complex and diverse than ever. Thus, multi-modal sentiment analysis is becoming a popular and promising research field that includes three main research interests: computer vision, natural language processing, and speech recognition. This work first introduces the background of this new technology and its progressive development around social media applications. Next, this paper represents a comprehensive overview of the advanced deep learning models used in multi-modal sentiment analysis. This review highlights several recurrent neural network (RNN) models and graphic neural network (GNN) models in recent studies. Here, we show that GNN based models can solve tasks in other domains such as physics, chemistry, and biology. After that, this work also includes several helpful summaries according to eight different papers. In the end, we illustrate an overall summary of our work. Some future research topics are also given with clear justifications.