Prediction of Public Opinion through Real-Time Monitoring using Multimodal Recurrent Neural Network

Abstract

There is growing use of technology and presence of people online globally. Electronic platforms have also become the avenue for expression of opinions on emerging issues by people. However, some of the posts or comments posted online could sometimes be negative with far reaching effects on the community. The use of machine learning algorithms through sentiment analysis offers means that could be used to mine data and analyse opinions that emanate from online to reach decisions or monitor ethical compliance. This study presents an approach that uses a Multimodal fusion with Recurrent Neural Networks (M-RNN) to predict opinions through decision making and real time monitoring. The dataset was trained using standard methods like a decision tree classifier, and the M-RNN model achieved an accuracy of roughly 82.80%. The training model's relative average error was close to 0.503% when using the M-RNN methodology, and the training cycle only needs to be repeated 250 times to achieve satisfactory results, a figure that is low when compared to other conventional methods.