Multimodal Sentiment Analysis using Late Fusion LSTM

Multimodal Sentiment Analysis is a proven method for effectively analysing and extracting information from data that belongs to several modalities. It helps to comprehend the sentiment that was expressed during communication. As a result, the primary problem is figuring out how to use the resources at hand to effectively assess the sentiment across all of the modalities while also overcoming the complexity of emotions in the thoughts that are presented. Numerous conventional techniques fall short in comprehending the global context that links the modalities. In this paper, we are proposing a Late Fusion Long Short-Term Memory Based architecture to predict sentiment using multimodal data. We have conducted the all the tests on the dataset CMU MOSI. The proposed architecture shows 71% accuracy on the mentioned dataset.