High-Accuracy Multilingual Opinion Mining using Multimodal Fusion and Deep Learning

Online shopping is an integral part of modern commerce, with digital purchases continually increasing. User-generated content on Web2 platforms offers valuable insights into products and services through critiques, recommendations, and suggestions. This paper presents a novel multimodal opinion mining system that integrates textual, visual, and acoustic data to enhance the detection of sentiments, including sarcasm and detailed user suggestions. The proposed system leverages advanced information retrieval techniques and deep learning tools to automatically identify suggestions within user-generated comments, aiming to improve organizational effectiveness in online commerce. By shifting from rule-based methods to deep learning approaches, the study demonstrates significant improvements in suggestion detection accuracy. The core innovation of the system lies in its multimodal fusion approach, employing a cross-modal target attention mechanism to integrate textual, visual, and acoustic features, thereby achieving superior performance in complex classification tasks. This study also addresses the critical challenges in opinion mining, such as the evolving nature of online platforms and the need for robust multilingual support. By advancing the field of suggestion extraction and sentiment analysis, this research sets a new benchmark for multimodal AI applications in e-commerce, offering significant implications for enhancing user experience and business decision-making. The High-Accuracy Multilingual Opinion Mining (HAMOM) algorithm developed in this study represents a significant step forward in understanding and leveraging user feedback for improved product and service development. Future research should continue to refine these multimodal strategies, ensuring that AI systems remain at the forefront of sentiment analysis and opinion mining advancements.