Multimodal Sentiment Analysis: A GRU Approach for Improved Accuracy

In the real world, data often comes in different forms, such as text, audio, and video, making it multi-modal. This poses a challenge for machine learning models, which need to be able to process and understand this type of data. One way to address this challenge is to use models that can handle multiple data types in a single model. This enables the model to learn from the diverse information offered by the various data types. This research explores various methods for sentiment analysis in a multi-modal context, incorporating text, images, and potentially audio data. The methods evaluated include lexicon-based methods, machine learning with text features, convolutional neural networks (CNNs) for text, and a proposed multi-modal GRU model. The objective of the multi-modal GRU model is to surpass existing methods by efficiently capturing and integrating sentiment information from various modalities. The evaluation of the model's capabilities and constraints encompasses multiple performance indicators, including accuracy, precision, recall, and F1-score. Our Multi-modal GRUs model achieved an accuracy rate of 85%, precision of 82%, recall of 80%, and an F1-score of 81%. These metrics indicate that the model adeptly captures the underlying patterns within the multimodal data.