Multimodal Sentiment Analysis (MSA) has advanced as a great approach to understanding human emotions. MSA utilized a significant number of different forms of modalities. The modalities include text (linguistics), visual, and auditory data. This paper reviews the state-of-the-art of MSA, focusing on fusion strategies in human emotion recognition. The purpose of MSA is to analyze data from different modes of modalities to analyze sentiment predictions. This paper discusses and highlights studies by other researchers for better insight into fusion techniques that can be deployed in MSA to address and understand the complexity of multimodal data. The aim is to enhance the model’s accuracy and robustness for better sentiment prediction. Despite the progress made, there are still significant opportunities and challenges in MSA, such as the complexity of the data to understand, noisy data and the consistency and correlation between the modalities. This also explains the insights into the prospective trajectories of MSA, where there is still a need to research and improve the current model to gain a more robust model and integrate new data sources.