Advanced Multimodal Sentiment Analysis Using Stacking for Enhanced Decision Support in Organizational Systems

This article discusses the development of an original multimodal sentiment analysis method designed to support a decision-making processes in organizational systems. With the development of technology and an increase in data volume, sentiment analysis methods have evolved from simple lexical approaches to complex models of machine and deep learning. Unlike already existing methods, the proposed model combines the advantages of several basic models optimized for different modalities, which allows for a significant increase in accuracy and stability of analysis. This research shows that the proposed approach provides up to ninety percent accuracy and up to ninety-six percent of F1-measure. These results surpass the outcomes of such well-known multimodal methods as Tensor Fusion Network and Multi-Attention Recurrent Network, especially when working with limited data. The proposed approach provides more accurate and effective analysis of an emotional coloring of data, which is a useful tool in making more informed decisions in various fields, including marketing, politics and economics.