Translation-Based Fusion for Multimodal Sentiment Analysis: Short Comprehensive Review

As social media continues to grow in popularity and multimedia content becomes increasingly prevalent, traditional text-based sentiment analysis methods may not capture the complete picture. Multimodal sentiment analysis (MSA) aims to overcome this limitation by incorporating multiple modes of data, such as text, images, and audio, to analyze and understand the sentiment expressed by individuals. This paper offers an extensive review of approaches employing translation-based fusion techniques to tackle challenges in Multimodal Sentiment Analysis. Firstly, it briefly outlines the fundamentals of MSA, including the MSA process, the importance of modalities, and the most widely used datasets. It also explains the existing categories of fusion methods in MSA. Subsequently, it details the architectures and key contributions of existing translation-based MSA techniques, and discusses the performance of these translation-based models on the two benchmark datasets, CMU-MOSI and CMU-MOSEI. It concludes by discussing challenges faced by multimodal sentiment analysis.