Exploring Multimodal Sentiment Analysis through Cartesian Product approach using BERT Embeddings and ResNet-50 encodings and comparing performance with pre-existing models

Multimodal Sentiment Analysis has become a popular interest of research in Machine Learning. It not only provides better results and context inference but also provides a good alternative for uni-modal analysis. The past few years have seen many new datasets for this task and state-of-the-art models dealing with the intermodular and intramodular dynamics of the task. This paper aims at exploring the pre-existing research on the topic of Multimodal Sentiment analysis and combining it with the idea of Sentence Embedding and Video encoding using state-of-the-art models, using the technique of classifying cartesian-product of extracted features, discussing experiments, and doing the qualitative analysis of the results.